On My Dance-Card…

“SynQuacer™ E-Series is a software development environment compliant with Linaro’s 96Boards open hardware specification. This has been built jointly by Socionext Inc., Linaro and GIGABYTE. It contains one “SC2A11”. “SC2A11” is a multi-core chip with 24 cores of ARM® Cortex-A53. It is possible to construct a high integrated and low power consumption server system. It is so highly efficient chip that it supports from cloud computing to edge computing, which is the key to IoT era.”
 
See DeveloperBox – 96Boards
This was announced a while ago and now it’s offered for sale. Unlike other cheaper setups this has 24 cores A-53 on a chip, socketed DDR4 RAM, comes from a consortium from China and Japan and comes in an ATX box. TDP for the CPU is 5W. One thing I don’t like about it is the price, $1230 USD. I won’t be rushing out to buy one but it’s definitely a good one to have on the list…

About Robert Pogson

I am a retired teacher in Canada. I taught in the subject areas where I have worked for almost forty years: maths, physics, chemistry and computers. I love hunting, fishing, picking berries and mushrooms, too.
This entry was posted in technology and tagged , , , , , , . Bookmark the permalink.

148 Responses to On My Dance-Card…

  1. An Out Of Phase Transistor says:

    Say, Doctor, did you actually set the argument so as to make Robert google this up?

    (Or are we seeing more of a “emergent gameplay” sort of a thing happening here, I wonder.)

  2. DrLoser wrote, “The Linux kernel does not even implement a locking strategy across its own kernel space data structures.”

    Utter nonsense. Linux had a “Big Kernel Lock” until 2011/2.6.39. Now it uses a more refined approach but the kernel can run on multiple cores simultaneously and locking is just fine. See also https://www.networkworld.com/article/2279761/software/kernel-space–the-big-kernel-lock-strikes-again.html

  3. DrLoser says:

    I’ll make this even simpler.

    The Linux kernel does not even implement a locking strategy across its own kernel space data structures.

    That may not be a sufficient argument for simpletons — but it should be.

  4. DrLoser says:

    Incidentally, and assuming that the kernel data structures are held in let’s say L3 for simplicity:

    Even without maxing out your 18 other cores, you’re going to have a ridiculously large amount of critical section contention, locking, etc, just between those two cores.

    You think not? Just examine the Linux kernel. It’s optimised to the max for a lot of things, but optimised for updating its own kernel-space data structures across more than one core?

    It ain’t.

    Sorry to disappoint you, Robert, you you clearly don’t know what you’re talking about here.

  5. DrLoser says:

    I can’t see what could possibly go wrong with this optimization strategy.

    Nobody was stupid enough to bite on this question (although oiaohm is certainly stupid enough, but couldn’t find an irrelevant cite), so I’ll just explain why this won’t work very well.

    Say you have two cores (for simplicity), both running the Linux kernel. Say you have eighteen more cores (not for simplicity, but to match the situation in the OP), all of which are maxed out.

    Here’s the thing, then. You have a high-priority process that needs to be swapped in. (You could in fact have twenty of them, just to point out the obvious.)

    One Linux kernel wants to free up a core for that. And you won’t believe this, but the other linux kernel also want’s to free up a core for that. But you’re short a free core, unless you want to swap one of the Linux kernels out.

    Precisely how do you think this is going to be achieved?

    It isn’t, is it? (I have to add this rider for people who can’t be bothered to think rationally. It isn’t.)

    Nope, sorry, one Linux kernel core to rule them all.

  6. DrLoser says:

    DrLoser Linux uses RCU [read-copy-update, because I prefer to use an explanatory name to a stupid acronym] on async communication to avoid locks.

    I love how you somehow manage to drag exotica out, Fifi, without having the faintest clue what you’re talking about. For the record, there are about 20,000 users of RCU right now. (I extrapolate from the cite. For all I know, everybody has given up on the thing.)

    “Linux” doesn’t use RCU, Fifi. A tiny proportion of Linux users give it a go.

    Oh, and to quote from that Wikipedia article?

    So the typical RCU update sequence goes something like the following:

    1) Ensure that all readers accessing RCU-protected data structures carry out their references from within an RCU read-side critical section.
    2) Remove pointers to a data structure, so that subsequent readers cannot gain a reference to it.
    3) Wait for a grace period to elapse, so that all previous readers (which might still have pointers to the data structure removed in the prior step) will have completed their RCU read-side critical sections.
    4) At this point, there cannot be any readers still holding references to the data structure, so it now may safely be reclaimed (e.g., freed).

    I don’t know about you, Fifi, but I count at least two locks there. The critical section in (1) and the reference count in (2). I suspect that the timer in (3) involves a lock too, so that would be three locks. Not “none.”

    Windows on the other hand has locks on async communication and this is why you can get lock failure while attempting to perform async on windows.

    It’s called “overlapping,” Fifi. It happens in the Windows kernel. There are no locks involved.

    This “lock failure” you speak of/ Care to quote an example?

  7. DrLoser says:

    No this is another case of being x86. Arm does not mandate that data in l1 and l2 caches also be in L3.

    Nothing does, you idiot. That’s not the argument. The argument is whether an OS kernel can sensibly work across cores without a common memory bridge, which in this case would be L3. (And even given that, the argument is that it would still be a stupid thing to do.)

    Not that it would be recommend on a 24 core system.

    OK, I take that back. You are not a 100% totally clueless imbecile. You are only a 23/24th totally clueless imbecile.

  8. oiaohm says:

    Ooh, ooh, I know! I know the answer! Windows async communication is non-blocking by definition and design, so … er … putting myself in oiaohm’s wretched little mind-space: no, I don’t.
    DrLoser you don’t know the answer.
    https://blogs.msdn.microsoft.com/jerrynixon/2013/02/04/walkthrough-locking-asynchronous-file-operations/

    Deaf Spy asked fifi who is not here so did not want a answer. Of course every time DeafSpy addresses a question to Fifi its normally completely wrong and you jumped in idiot boots and all DrLoser.

    DrLoser Linux uses RCU on async communication to avoid locks. Windows on the other hand has locks on async communication and this is why you can get lock failure while attempting to perform async on windows. Linux has real async communication and windows has pretend async communication this does make a big difference as number of cores increase.

    All the data structures, as we are led to believe by people who don’t know better or indeed anything at all on the subject, will still reside in L3 cache.
    No this is another case of being x86. Arm does not mandate that data in l1 and l2 caches also be in L3. One of the fun features of arm data to and from l2 cache can go to mmu without going though l3 because you can disable l3. Not that it would be recommend on a 24 core system.

  9. DrLoser says:

    The nice thing about the linux kernel occupying more than one core/cpu, btw, is that it would be able to swap itself out.

    All the data structures, as we are led to believe by people who don’t know better or indeed anything at all on the subject, will still reside in L3 cache.

    I can’t see what could possibly go wrong with this optimization strategy. Apart from, of course, the obvious fact that the Linux kernel doesn’t do it.

  10. DrLoser says:

    Configuration parameters, Robert. At your leisure, if you will.

  11. DrLoser says:

    Fifi, a simple question for you. In Windows, when you start an asynchronous IO operation, it there a thread (in whatever space) that waits for the operation to complete?

    Ooh, ooh, I know! I know the answer! Windows async communication is non-blocking by definition and design, so … er … putting myself in oiaohm’s wretched little mind-space: no, I don’t.

    How extraordinary that oiaohm has failed to respond to this. I wonder why?

  12. Deaf Spy says:

    Fifi, a simple question for you. In Windows, when you start an asynchronous IO operation, it there a thread (in whatever space) that waits for the operation to complete?

    Entertain us, Fifi.

  13. DrLoser says:

    Configuration parameters, Robert. At your leisure, if you will.

  14. DrLoser says:

    But let’s just assume that I, as either Linus or Microsoft or indeed anybody else, had some perverse desire to “fake” a pre-emptive scheduler. And for our current purposes, let’s ignore Robert’ s rather interesting notion that you can spread the kernel over more than one cpu/core, with goodies aplenty.

    Let’s say I want to “fake” a pre-emptive scheduler.

    Wouldn’t it be easier just to implement a pre-emptive scheduler?

    Come to think of it, those flags on the Moon? Faked, in Arizona. The twin towers falling? All down to three weeks’s worth of installing explosive charges on the structural members.

    I can see where this is coming from, and it’s not just rampant stupidity.

    Anything and everything that does me harm is a conspiracy, obviously.

    (Or, alternatively, Microsoft Windows has had a pre-emptive scheduler since NT was invented in the first place. But that’s far too boring for people with metal spikes in their head.)

  15. DrLoser says:

    Let me expand upon that nonsensical claim that Linux somehow “spreads async I/O” across more than one thread, btw.

    And let’s assume that Linux has proper async I/O, which is questionable. And let’s assume that it somehow takes the, if you like, “request” from one thread and, with magic pixie dust, finds the best possible other thread to hand the “response” to.

    Imagine, if you will, the request thread. Let’s say it’s a web server, or client, with ~20MB of state hanging around. Some of that is going to be useful, some of it is going to be crap from ad servers.

    Now imagine the response thread. Hey! I woke up! Same ~20MB of state. You don’t even know which bit is crap, because you are a brand new thread.

    There goes, not only your L1 and L2 cache (pretty much the reason for async I/O in the first place), but also your L3 cache.

    I’d like to pour scorn on Linux for doing this — but actually it’s a mature system.

    So it doesn’t.

  16. DrLoser says:

    Still, the rodent can prove me wrong

    There’s a first for everything, I suppose. But “proof” is not in oiaohm’s vocabulary.

    Besides, he’d only misspell it as “poof.”

  17. An Out Of Phase Transistor says:

    Oh, were you trying to accost Hammie (and, in pasing, Robert)?

    Carry on then.

  18. An Out Of Phase Transistor says:

    Meh, it’s out of line, I blame Robert for this (entire blog)!!

  19. An Out Of Phase Transistor says:

    Well, Doctor, I must ask:

    HOW DARE YOU BE THE JUDGE OF THINGS ONLY HAMSTERS ARE MEANT TO JUDGE!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

    But, in all sincerity, I must admit that passing as a hamster is as difficult for me as it is for Hammie to do the opposite (if one dares to posit the issue as a binary opposition, which it may not be, could be we are looking at a spectrum disorder here – a hamster to rodent spectrum that is).

    Still, the rodent can prove me wrong, all he (they/it/…) need(s) to do is post a half legible and possibly even meaningful sentence (just one Hammie, you can do it; google for it, then use copy-paste, c’mon).

  20. DrLoser says:

    Anyway. About those “configuration parameters,” Robert.

    You are an honest man and I trust you to research them. Let us know the results of your research.

  21. DrLoser says:

    Sigh. One more blatant imbecility.

    windows use 1 thread however it fake premptise sceduler. PeekMessageW!=0 it not yield process however of PeekMessageW=0 it yield process however on timer

    So, you are saying that the OS kernel — any OS kernel — faithfully checks the zero flag in assembler on each and every process/thread in a tight look, and acts in a different way depending upon whether that zero flag is 1 or 0?

    Call me a riotous maniac with a propensity to hallucinogenic dreams, but somehow, I can’t quite bring myself to believe that this is true.

  22. DrLoser says:

    DrLoser Linux has better asyncronus IO hence it use >1 threads.

    Let’s ignore the appalling spelling. Let’s ignore the weirdo concept of spreading asynchronous I/O across more than one threads. (What you gain in asynchronous efficiency, you lose in thread interlocks and scheduling. Funnily enough, this is the entire point about thread affinity, a subject you brought up in the first place.)

    No, Linux does not have “better asynchronous I/O” In fact, unless things have changed since about 3.4 or so, I don’t believe it has any asynchronous I/O at all. This is, I admit, a matter where I may not be up-to-date, in which case I welcome a cite to that effect.

  23. DrLoser says:

    Hey look, I, too, can try the Hammie-conversation

    Too many commas, too much sense. Needs more (or possibly less) work.

  24. DrLoser says:

    windows use 1 thread however it fake premptise sceduler. PeekMessageW!=0 it not yield process however of PeekMessageW=0 it yield process however on timer

    It’s a tight loop, young man. It isn’t quite a spin-lock, but it’s close enough.

    There is no faking here at all: there is simply a pre-emptive scheduler. Leaving aside* the possibility of high and low priorities, each thread is allotted a time slice. You can loop tightly around whatever idiotic “ping” you want, and neither the Windows kernel nor the Linux kernel nor any other sane kernel will evict you until the time-slice is exhausted. Why should they? How could they possibly know?

    * Bringing back process priorities: some OSes will still wait on the end of the time-slice and then pick the next in line on the basis of process priority. Others, and we normally refer to them as “soft real-time schedulers,” will interrupt a time-slice on an algorithm based upon relative process priority.

    I do hope this has made things a little clearer to you, little fellow. I doubt it, but hope springs eternal.

    Oh, and btw, a process/thread waiting on a timer is in the idle state. Once the timer expires, it enters the wait state. After that it is up to the scheduler to pick a process off the wait list and promote it to the run list.

    I really don’t think I can make this any simpler. It’s actually so easy to explain that an eight year old child could understand it. Have a lollipop!

  25. joepeatf says:

    Tell us again how Peek MessageW somehow “blocks” a pre-emptive scheduler?

    For extra goodies, do please analyse the name of the function. Peek means peek. Ya wanna put it in a tight loop? Nobody is going to stop you being a total moron.
    DrLoser I did mistake again. I tested code it does not block sceduler however microsoft windows use timer.

    windows use 1 thread however it fake premptise sceduler. PeekMessageW!=0 it not yield process however of PeekMessageW=0 it yield process however on timer

    windows process has 1 timer..it not use thread however it use Rdtsd QueryPerformanceCounter I believe. windows Xp is slow hence it use 500ms timer however windows 7 fast hence it use 50ms timer.

    PeekMessageW save Rdtsd QueryPerformanceCounter last time

    now=QueryPerformanceCounter
    a=now-last
    if a>500{
    yield
    }
    now=last

    evidence

    Why windows Xp not undraw moved window (not responding) ?

  26. An Out Of Phase Transistor says:

    In fact, Hammie, Windows uses all threads in existence, that’s why there are no left for Linux, hence all of the problems you’ve been observing.

    The aggravation you feel is normal, the threading model is activating nearby bluetooth devices, please remain calm.

    (Hey look, I, too, can try the Hammie-conversation; if it isn’t as good as the original, just wait for the inevitable response >:).)

  27. An Out Of Phase Transistor says:

    Sez DrLoser:

    Back in those days, [Fifi] somehow believed that Out of phase transistor and I were identical twins.

    He did? I am as much amaze as macaroni is cheeze-pasta (trying to relate to the Hamster here, if you don’t mind).

    Also, we weren’t? The shock, the horror, I never knew…

    Ha ha ha…

  28. joepeatf says:

    Fascinating. You claim that the Ubuntu package repository stops you from writing such a simple thing?
    Ubuntu package repository not stops you from writing .however distro maintainer ban virus on Aptget repository.

    Appimage is Linux less security..Fred gets virus on Linux however Aptget repository is safe

  29. joepeatf says:

    Fascinating. If only we knew what “Linux app” you are blithering on about.

    Let me refresh your memory. Qt, gcc, two threads, on ginormous frickin stupid unrealistic giant all-screen button, plus another process (fork it if you want) that works off a timer and displays a “top-level” dialog box.=ban “Linux app”
    DrLoser Linux app runs on Ubuntu 18 however distro maintainer bans virus on Aptget repository. It not run on Fedora Gentoo Arch also. Linux uses premptise sceduler also. DrLoser Linux has better asyncronus IO hence it use >1 threads. macrosoft windows need GetMessageW always hence it use 1 threads.

  30. DrLoser says:

    These “configuration parameters,” Robert.

    We’re waiting. And we won’t let Fifi gibberish get in the way of us waiting.

    What are they? After all, one of your minor hobbies is to spin up a build of the Linux kernel every now and again. I presume you’ve checked out the configuration parameters that specify the number of cores/cpus available to the Linux kernel.

    I mean, those are probably quite important, when you’re tuning your own version of the kernel.

  31. DrLoser says:

    DrLoser Linux app is ban in Ubuntu package repository. Linux app need be QA by Ubuntu’s distro maintainer.

    Fascinating. If only we knew what “Linux app” you are blithering on about.

    Let me refresh your memory. Qt, gcc, two threads, on ginormous frickin stupid unrealistic giant all-screen button, plus another process (fork it if you want) that works off a timer and displays a “top-level” dialog box.

    Fascinating. You claim that the Ubuntu package repository stops you from writing such a simple thing?

    Fascinating. It’s probably because Ubuntu forbids you from writing your own “app” in C unless you use a GUI, just like Microsoft insists that you do, except it doesn’t.

    Take your meds, Fifi. Now.

  32. DrLoser says:

    But it’s absolutely lovely that whichever Fifi it is admits, for once, to an error. I can’t honestly remember this ever happening. The poor little thing must be feeling the pressure of having reality hammered into him.

    I did mistake DrLoser, PeekMessageW block sceduler.

    Your first mistake, Fifi, was to assume that your idiot copy-pasta “virus” program could somehow block the Windows scheduler. One would have thought that even a minimally-alert mental deficient would have stopped right there. But, I will hold my hand up. This is (I think) the very first time you have ever admitted to making a mistake. You’re making progress!

    Not very much though. Funguses grow faster than your progress in CS.

    Tell us again how Peek MessageW somehow “blocks” a pre-emptive scheduler?

    For extra goodies, do please analyse the name of the function. Peek means peek. Ya wanna put it in a tight loop? Nobody is going to stop you being a total moron.

    For even extra goodies, do please apologise for the equivalent *nix version, which is (in C):

    while (1) do {
    select (0, NULL, NULL, NULL, NULL)
    }

    Same completely moronic programming. Different OS.

  33. DrLoser says:

    C complier need a lot of lines hence microsoft cheat.

    Fascinating.

    microsoft use C89 only hence need less line

    Fascinating.

    less optimize also

    Um, how does the number of “lines” correlate to the algorithms for optimization?

    Fascinating.

    gcc optimize better then microsfot..clang also

    Fascinating, yet unproven by you. And in all honesty nobody cares.

    Take your meds, Fifi Mk II. Go back to being Fifi Mk I. You’re both imbeciles, but at least one of you seems to be vaguely hominid.

  34. DrLoser says:

    while(PeekMessageW) is spin lock hence it block cooperate sceduler.

    That would be a very good reason not to call PeekMessage inside a tight while-loop, wouldn’t it? It’s not like this spin-lockk comes out of nowhere.

    And I agree, it would block a “cooperate sceduler.” The only problem with your masterful thesis is that there is no co-operative scheduler involved.

    Remind me again, which one of you uses commas, and which one does not? Also, which one of you spells “compiler” habitually as “complier,” and which one does not?

    Looking down your recent posts, Fifis, there is no clear difference. Perhaps one of you could put your hand up and say “I! I am Spartacus!”

  35. DrLoser says:

    Fred has virus.

    What a very interesting observation. Not in any way relevant, but interesting nonetheless.

    I would suggest, Fifi, that the next time you meet “Fred” under the lamp-post, you make sure that no bodily fluids are exchanged. Manual satisfaction is paid at a lesser rate, I am led to believe, but at least you can disinfect your hands afterwards.

  36. joepeatf says:

    But perhaps Windows is “cheating” in some way. Perhaps it has “secrit codez” that catch this “flaw,” what with copy-pasta-kiddies having used it for twenty years.

    Let’s try the same experiment again, but this time using Qt and gcc. Nice, portable, stuff. Works on both Windows XP and on the Linux of your choice.

    Without the second process (dialog box, nowt much else), you will find that both Windows and Linux end up with a massive great fat pointless screen-gobbling button.

    With the second process, you will find that the scheduler is happily toiling away in the background.
    DrLoser Linux app is ban in Ubuntu package repository. Linux app need be QA by Ubuntu’s distro maintainer. Virus hacker use Launchpad.. Launchpad is less security then Ubuntu package repository.

    However windows app upload to phishing site..Fred download fake Google’s Chrome..Fred has virus.

  37. joepeatf says:

    Set yourself up two processes. One of them, on a timer, pops up a dialog box every ten seconds or so, as a “top window.” The other one does this nifty little copy-pasta (perhaps it’s macaroni?) trick.

    I think you’ll find that, when the timer goes off, a dialog box will appear “as if by magic” on top of the pointless great big fat screen-gobbling button. The reason for this is that Windows XP is not a co-operatively scheduled system: it’s possible to pre-empt one thread with another.
    I did mistake DrLoser, PeekMessageW block sceduler..

    GetMessageW make process -> wait state. It wait however if Fred touchs windows app it make process -> run state.. Withal if Fred download file in internet explorer, doned download go to GetMessageW queue. However windows app is single thread, Linux Gui app use >1 thread hence it is more reponseve.. it use >1 Cpu core.

    while(PeekMessageW) is spin lock hence it block cooperate sceduler.

  38. joepeatf says:

    Not that this is relevant in any way, but let’s explore your fantasy further. You are proposing that all C compilers should be, what, a single line long? Or perhaps you have a secret definition of the phrase “lot of lines,” which allows a conformant non-bloat C compiler to be maybe as much as 24 lines long? (More than that and simply taking your shoes and socks off won’t help. I am assuming that you have particularly gifted tootsies.)=straw man arguement.
    Caption obvious C complier need a lot of lines howeverr windows app need a lot of line hence windows is a lot of bloat. Linux Gui app need less line, gtk is better then windows.

    C complier need a lot of lines hence microsoft cheat. microsoft use C89 only hence need less line..less optimize also..gcc optimize better then microsfot..clang also

  39. DrLoser says:

    Of course, Robert, you may have come up with those “configuration parameters” you mentioned earlier, and upon which I challenged you, that can spread the Linux kernel over more than one cpu/core.

    I mean, the fact that you haven’t come up with them so far doesn’t mean that they don’t exist. You might just be holding them in rserve, as a final triumphant argument.

    Which is not nice, Robert. Don’t rub my face in it. I’ve put a lot of effort into this. Honour that effort, and tell me what these nonexistent configuration parameters might be.

    Otherwise, you really don’t have any standing in this discussion at all, do you?

  40. DrLoser says:

    glibc sets some bits associated with the processes. The kernel’s scheduler uses those bits to choose a core to run the process. It’s all good.

    Correct-ish. As pointed out, the glibc API doesn’t really do anything relevant to the kernel, apart from making syscalls. Who knows? It might set those bits for its own convenience. But those bits are not relevant to the bits set within the kernel for the same purpose.

    Therefore meaningless in context. Or maybe not. The kernel’s scheduler uses those bits to choose a core… … which implies what, precisely, about the location of the kernel, Robert?

    Do save yourself the embarrassment. I have a couple more levels to go before I run out of ways of proving how completely inaccurate the idea of spreading the Linux kernel over more than one core is.

    Seriously.

  41. DrLoser says:

    But perhaps Windows is “cheating” in some way. Perhaps it has “secrit codez” that catch this “flaw,” what with copy-pasta-kiddies having used it for twenty years.

    Let’s try the same experiment again, but this time using Qt and gcc. Nice, portable, stuff. Works on both Windows XP and on the Linux of your choice.

    Without the second process (dialog box, nowt much else), you will find that both Windows and Linux end up with a massive great fat pointless screen-gobbling button.

    With the second process, you will find that the scheduler is happily toiling away in the background.

    I look forward to the next instalment of “I need a clue-bat! joepeatf demonstrates his complete and utter ignorance for fun and profit!”

  42. DrLoser wrote, “glibc resides in user space and is no sort of an indication whatsoever of which core the kernel is sitting on.”

    glibc sets some bits associated with the processes. The kernel’s scheduler uses those bits to choose a core to run the process. It’s all good. See https://stackoverflow.com/questions/766395/how-does-sched-setaffinity-work

  43. DrLoser says:

    It’s a futile gesture, I know, but I’ll expand upon Deaf Spy’s excellent analysis of this “virus.” Incidentally, this ninny effort has been around for at least twenty years. I can’t place where joepeatf has copy-pasta’ed it (possibly with added linguine, though definitely not in this case with either ravioli or lasagne), but it’s rather sad when you can’t even make up your own stupid example but have to crib it from somewhere else, isn’t it?

    And as an aside, as Deaf Spy points out, this is in no way a virus. What it is, is a combination of two things (as well as cribbing):
    1) PEBCAK — do not let joepeatf anywhere near a computer, under any circumstances.
    2) What Raymond Chen describes as “It rather involved being on the right side of this airtight hatchway.” In other words, the person who shot himself in the foot was you. Directly.

    But let’s test this Miracule Windows Denial Fraudulene a bit further. To quote Deaf Spy:

    Let me know explain it to you, dimwit. Your application:
    1) doesn’t stop the scheduler. It doesn’t effect the scheduler in anyway.

    Perhaps Deaf Spy is wrong. Perhaps, despite the manifold other deficiencies of this absurdity, it really does affect, or stop, the scheduler.

    Set yourself up two processes. One of them, on a timer, pops up a dialog box every ten seconds or so, as a “top window.” The other one does this nifty little copy-pasta (perhaps it’s macaroni?) trick.

    I think you’ll find that, when the timer goes off, a dialog box will appear “as if by magic” on top of the pointless great big fat screen-gobbling button. The reason for this is that Windows XP is not a co-operatively scheduled system: it’s possible to pre-empt one thread with another.

    Which is probably too complex an idea to force into joepeatf’s head. So let’s just say that this elementary demo would prove beyond all doubt that the scheduler is still doing its job.

  44. DrLoser says:

    DrLoser microsoft windows C is a lot of lines hence microsoft windows is bloat OS

    Not that this is relevant in any way, but let’s explore your fantasy further. You are proposing that all C compilers should be, what, a single line long? Or perhaps you have a secret definition of the phrase “lot of lines,” which allows a conformant non-bloat C compiler to be maybe as much as 24 lines long? (More than that and simply taking your shoes and socks off won’t help. I am assuming that you have particularly gifted tootsies.)

    We’ve all been using C compilers wrong these many years, haven’t we? We should follow your brilliant lead and start off by checking the number of lines (for bloat), and then we should compile and link the compiler so that we can actually use it to,, you know, compile stuff.

    Strikes me as a singularly bloated work-flow, but each to his own.

    windows need GUI also.

    Let me introduce you to the notion of Windows Services. It is very rare (and I would say very poor practise in general) to write a Windows Service that “need” a GUI.

    Wrong again. Stop playing with yourself.

  45. DrLoser says:

    DrLoser pthread_setaffinity_np is kernel function hence it use all core however it need code be in.

    Well, actually, it’s a glibc function, as can plainly be seen by reading the man page. You do know what a man page is, don’t you, joepeatf? Presumably not.

    You may remember that I urged Robert to explore the notion of glibc, I imagine. Well, it turns out that Robert is far too sensible and knowledgeable to make this elementary mistake. You, on the other hand …

    A couple of things here. glibc resides in user space and is no sort of an indication whatsoever of which core the kernel is sitting on. Further to that, pthread_setaffinity_np isn’t even a syscall in Linux. Which means it doesn’t even talk directly to the kernel.

    I am therefore not going to bother analysing your gibberish code, because it has no bearing on the matter.

    Oh, and since pthread_setaffinity_np is intended specifically to lock a given thread to a subset of cpus/cores, it almost certainly doesn’t use “all core,” whatever that is intended to mean. I suppose you could set every bit in the affinity mask, but then what would be the point of making the call in the first place?

    Could you be more wrong?

  46. Deaf Spy says:

    DrLoser I made virus. It stops sceduler hence windows is freeze. I use windows xp, it lock windows

    Bwa-ha-ha-ha

    Virus, oh, dear, oh sancta simplicitas! That was a good one, Fifi, I still shake with laughter. Fifi II, you are the Fifi I version without meds, dear.

    What you did, you dimwit, is to create a top-level window the shape of a button that covers all other windows (except other top-levels that appear after it).

    Let me know explain it to you, dimwit. Your application:
    1) doesn’t stop the scheduler. It doesn’t effect the scheduler in anyway.
    2) can be stopped using TaskManager.
    3) is not a virus – it doesn’t replicate itself, it is a pitiful trick that only a pimple-ridden 12-year old can be happy with.
    4) affects only the interaction of the current user, while the rest of the system and apps are happily running.

    You are clueless, Fifi.

  47. joepeatf says:

    Perhaps it’s the operating system to blame after all. You should try Windows, which is far more efficient at this sort of thing.
    microsoft windows is not efficient C complier is lees optimize C89 also. micrasoft windows is coperate sceduler hence coder need sceduler codes.

    DrLoser does not know windows. windows C is not standard it has Posix however it not has all Posix. windows use notstandard Api. ll

    DrLoser I made virus. It stops sceduler hence windows is freeze. I use windows xp, it lock windows

    #include <windows.h>
    int wWinMain(HINSTANCE A,HINSTANCE B,LPWSTR C,INT D){
    HWND E=CreateWindowExW(WS_EX_TOPMOST,L”BUTTON”,L”NO XBUTTIN HENCE microsoft=LOCKIN”,WS_CAPTION,314,159,265,358,NULL,NULL,A,NULL);
    ShowWindow(E,SW_MAXIMIZE);
    MSG F;
    while(GetMessageW(&F,NULL,NULL,NULL)){
    TranslateMessage(&F);
    DispatchMessage(&F);
    }
    }

    DrLoser microsoft windows C is a lot of lines hence microsoft windows is bloat OS. windows need GUI also..

    windows.h is notstandard Api bloat also. Its all capital letter almost

  48. joepeatf says:

    Symmetric Multi-Processing refers to user space, not to kernel space.
    DrLoser Symmetric Multi-Processing is kernel space also. Linux sceduler is all cores not posible 1 scedules all of 24 cores. Sceduler need be less work hence use all cores. Example Bob client use core1 Fred client use core2 however Fred client use a lot of processor fork hence core2 has sceduler. Bob client need less sceduler hence Sally client use core1 also. Essense kernel use Symmetric Multi-Processing

    void main(){
    int thinclts[24],all=sizeof thinclts,bob=thinclts[0],a;
    memset(thinclts,0,all);
    bob=fork();
    memset(thinclts,bob,all);
    for(a=0;a<all;a=a+1){
    int x=thinclts[a];
    pthread_setaffinity_np(x,24,NULL);
    not parent process
    if(x=0){
    system(“echo hello bob”);
    system(“echo THIN CLIENT!!!”);
    }
    }
    system(“pause”);
    }

    DrLoser pthread_setaffinity_np is kernel function hence it use all core however it need code be in.

  49. DrLoser says:

    I know the Beast is wheezy (a little Debian joke there for you), but if it really needs to spread a database lookup for a recipe, or “poking a little data,” across all four CPUs, then it’s in trouble, Robert. It’s hard to imagine any advantage to spreading this thin gruel of a load across more than about 10% of a single CPU.

    Perhaps it’s the operating system to blame after all. You should try Windows, which is far more efficient at this sort of thing.

  50. DrLoser says:

    Let me refresh you memory as to the context of your claim, just in case you try to weasel out and pretend that you were not talking about the Linux kernel:

    DrLoser wrote, “That’s a very generous nine cores, Robert. Why waste money on the other eleven, long hair or not?”

    … BTW, the OS uses way more than one core if available.

    I think we can both agree that “nine” is substantially greater than “one.” You are therefore responding to an admission on my part that (in theory) you might be able to use nine cores for user-space processing.

    In this specific context, there is clearly little or no point to you talking about “way more than one core” unless you are specifically talking about the OS, ie the kernel.

  51. DrLoser says:

    I’ve never made that claim.

    Yes you have, Robert. What part of “BTW The OS uses way more than one core if available” do you not understand? One would hope all of it, since you were the one who said it. That is your claim and you cannot honestly deny it.

    Linux does use multiple cores if available and configured to do that.

    For the kernel? Does it? What are the relevant configuration variables? It doesn’t, does it?

    It’s a little thing called SMP and it’s been around since the ’90s.

    Symmetric Multi-Processing refers to user space, not to kernel space.

    When I look for a recipe or poke a little data all four CPUs get some work.

    Work done in user space at the behest of the kernel, running on a single cpu/core, Robert. Have you lost your mind? Why not do as I suggest, and quote actual statistics from Beast? Might it just be that you know they will disprove this silly notion that the Linux kernel runs on more than one cpu/core?

  52. DrLoser wrote, “Come back and tell us when the Beast uses more than even 50% of a cpu/core running the OS. You won’t, because it won’t.”

    I’ve never made that claim. Linux does use multiple cores if available and configured to do that. It’s a little thing called SMP and it’s been around since the ’90s. When I look for a recipe or poke a little data all four CPUs get some work.

  53. DrLoser wrote, “The problem is, Robert, you don’t. And you never have.”

    Nonsense. I’ve run a whole school for a day at Easterville using Beast. I’ve several times run a whole lab of 24-30 teenagers. Believe me, they demand service. Here at our home we have 3 desktop-like seats and two smartphones running through Beast all day long and there are four human users and many processes/daemons too. “Load average” on Beast was 0.28 over the last 15 minutes. It gets to ~7 quite comfortably when the house is full of guests/visitors.

  54. DrLoser says:

    Need is a relative thing. If you have ten users running heavy I/O on 6 hard drives and a couple of Ethernet ports, it’s just more efficient to have some stuff run on one core and something else run on another.

    The problem is, Robert, you don’t. And you never have. That ridiculous list of yours is proof of this: it doesn’t include anything to do with resource usage, just a list of largely idle processes.

    But, as you say. Need is a relative thing. It strikes me that Beast is pretty decent at fulfilling your needs, considering that it is, what, seven years old? So let’s extrapolate this over seven years of development, using your old favourite, Moore’s Law.

    You’ve got six cpu (I think) running at an energy-burning 3.x GHz. This suits your computing needs, but not your energy bill needs, and anyway it’s about to die (God bless the Beast!).

    Call that 20 core-GHz in 2011 terms, for convenience. Feel free to adjust the base.

    Now, as you may possibly be aware, in the unlikely case that you bothered to read the copious on-line information about Moore’s Law, that law states that the amount of raw computing power doubles every 18 months. We’re looking at, say, 84 months here. Which is 2^4.5. Which is ~20. Let’s do the maths!

    20/20 := 1. Congratulations, Robert: using Moore’s Law, you can replace the Beast with a single core!

    But hey, you need a bit of wiggle room. Sacrifice performance for energy efficiency. Add a bit of extra oomph for the hordes that descend with their thin clients. That sort of thing.

    Looks to me that what you actually need is an Intel 8th gen 4 core NUC. Coming up later this year, probably priced at 2/3 what this piece of dreck would cost you. If you bought it. Which you won’t. Because you’re too mean.

    Go for at least a 1.6 GHz clock speed, man — otherwise your OS will bottle out at inappropriate times.

  55. DrLoser says:

    Both Deaf Spy and I would like to explore this wondrous claim further, Robert. And I can show you the tools to do so.

    BTW, the OS uses way more than one core if available.

    There are architectures where this is feasible, Robert. One that springs to mind is QNX, which you can extend to various forms of Mach if you like. They work on message-passing. They’re optimised for it.

    So, for example, you could have your scheduler on one core, your network services on another core, your GPU access on another core, your disk access on another core (one per disk, if you want), and your I/O on another core.

    This is called a “micro-kernel architecture,” Robert. It’s innefficient, but it’s worth it for trade-offs like resiliency and narrow attack surfaces.

    Unfortunately, you are not aiming for this architecture. You are aiming for a Linux architecture which (and this is common across all popular OSes) is anything but a microkernel.

    So, let’s just consider the scheduler. Helps if it’s on one core. Can you imagine how a scheduler would work across more than one core? You can stick all the data structures in L3, sure, but you’re still going to go through an awful lot of interlocks … which aren’t going to give you any benefit whatsoever.

    (At this point I should parenthetically point out, Robert, that I have been working with multi-processor servers for going on 30 years, commercially. Which is going on for 30 years more commercial experience than you have. Incidentally, the server standard is “<80% cpu usage for the OS, or you're a dead man thrashing.")

    But, back to comparisons between your current workload (Beast) and your notional, but you'll never buy one because $1,250 is far too rich for your blood, 24 core replacement. Let's see what we can glean, shall we?

    Well, the two have similar architectures. And they both (presumably) use the Debian kernel. You're probably not going to magic up a multi-core OS on the second, unless you can magic one up on the first. I assume Debian is up to the latest standards in graphical resource usage — say, Win 98 — and that you can therefore measure these things. (I told you I’d offer up helpful advice.)

    Go on, Robert. Measure them. Come back and tell us when the Beast uses more than even 50% of a cpu/core running the OS. You won’t, because it won’t.

    I sense something in the air, however. I sense the magic word glibc,, or something similar.

    Go ahead and make that argument, Robert. I implore you.

    Show Deaf Spy and me your Petticoats of Ignorance.

  56. Deaf Spy wrote, “If the OS does need more than one core to do its chores, then you are in very, very serious trouble.”

    Need is a relative thing. If you have ten users running heavy I/O on 6 hard drives and a couple of Ethernet ports, it’s just more efficient to have some stuff run on one core and something else run on another. That’s what a true multi-user/multi-tasking OS will do. See Beast:
    cat /proc/interrupts
    CPU0 CPU1 CPU2 CPU3
    0: 44 0 0 0 IO-APIC 2-edge timer
    1: 0 1 21 13142 IO-APIC 1-edge i8042
    7: 1 0 0 0 IO-APIC 7-edge parport0
    8: 0 0 0 1 IO-APIC 8-edge rtc0
    9: 0 0 0 0 IO-APIC 9-fasteoi acpi
    14: 0 0 0 0 IO-APIC 14-edge pata_atiixp
    15: 0 0 37 35099 IO-APIC 15-edge pata_atiixp
    16: 0 521 158972 53613 IO-APIC 16-fasteoi ohci_hcd:usb3, ohci_hcd:usb4, snd_hda_intel:card0
    17: 0 0 0 2 IO-APIC 17-fasteoi ehci_hcd:usb1
    18: 0 0 1543566 1468351 IO-APIC 18-fasteoi radeon, ohci_hcd:usb5, ohci_hcd:usb6, ohci_hcd:usb7
    19: 0 0 3 287 IO-APIC 19-fasteoi ehci_hcd:usb2
    22: 0 20 243628 110404 IO-APIC 22-fasteoi ahci[0000:00:11.0]
    27: 0 0 3 268 PCI-MSI 1048576-edge ahci[0000:02:00.0]


    Why on Earth should an OS not use all available resources to get the job done?

  57. DrLoser says:

    You have to pay subscription to access the on-line form “The Macquarie dictionary”. Yes it is the official dictionary of Australia and is the one used in court cases here.

    Which is wonderful. But I’m not about to pay a subscription, just to go down one of your inadequately explained rabbit-holes.

    There is such a thing as “fair usage,” you know. You are permitted to quote the specific text of these things. You are, of course, not going to do this, because you know you are wrong.

  58. DrLoser says:

    Proper noun has capitalization. But general and specific nouns don’t have to have capitalization unless they happen to be a proper noun in the context they are used or some other rule like sentence capitalization says they have to be.

    They’re called “common” nouns, oiaohm. They are not called “general or specific” nouns. There is no such grammatical term.

    There are also “indefinite” nouns, “abstract” nouns, and “concrete” nouns. But to repeat, there are no “general” nouns and there are no “specific” nouns.

    I don’t know how much your mother paid your kindergarten teacher, but whatever it was, I think you’re due a refund.

  59. Deaf Spy says:

    BTW, the OS uses way more than one core if available.

    If the OS does need more than one core to do its chores, then you are in very, very serious trouble. Do you understand that kernel time translates to?

  60. Deaf Spy says:

    Uh, multiple users, multiple processes, stuff like that.

    Uh, what multiple users, except yourself and your wife? And, what would be these so many mythical processes that are not idling in 99% of their time?

    There are times when 30 people use the network for instance, during parties

    Uh, you mean 30 people browse your db with recipies? Or do you mean 30 people connect to your pathetic router? I highly doubt your wireless router can sustain 30 active connections, and I definitely don’t think these 30 user your “services”.

    , or if I build a kernel I could use even more.

    Have you heard of project dependencies?

    Beast is not about throughput alone but service/responsiveness.

    One of the prerequisite for responsiveness is exactly throughput.

  61. oiaohm says:

    DrLoser how many cores will the following use gimp 3.10, firefox, chrome, blender and a growing list of applications use. Yes way more than 1.
    https://wiki.postgresql.org/wiki/Parallel_Query_Execution
    Database is critical postgresql can eat up all 24 cores by itself at times even running word-press. As we are getting more and more multi threaded applications be you running thin clients or not more cores have advantages. More and more of these programs are in the 90%+ multi-threaded class.

    Of course its unlikely that 24 cores will be always running at full load so being power effective under light load is what makes this number of cores viable.

    https://www.ntid.rit.edu/sea/processes/articles/grammatical/types/specificgeneral
    A definition of “general noun,” please. Not a recognised grammatical form, I think.
    Of course DrLoser does not know basic grammatical term that is globally part of English languages . This makes me suspect DrLoser has the UK monkey see monkey do form of English training this makes him totally unable to cope with the global variations in English languages.

    Proper noun has capitalization. But general and specific nouns don’t have to have capitalization unless they happen to be a proper noun in the context they are used or some other rule like sentence capitalization says they have to be. Basically all nouns types other than proper don’t have mandatory capitalization.

    If you had got the dictionary as I had told grece would would have notice the warped entry of lower case english marked general noun and plural for English languages so writing “english”= “English languages” by Australian dictionary as one of the defines. So I was using a local way of writing it. So grece would have been right to ask be to use a more globally know form but what I wrote was correct for my nationality.

    Really DrLoser you want to say giberish because if you had looked you would have worked out that there are truly national differences with English languages. Please note languages is the plural.

    Also you find what I have written in Australian authored books on English Languages using english the way I just did. So its not me being a special snowflake.

    To repeat, “You forgot to sprinkle in the commas, oiaohm. So, not quite the distinguishing mark you claim.

    I will repeat I never use commas never have. So its not me adding them. There are other differences in word selection. Also its not just commas DrLoser how often do I miss putting in a full stop at end of paragraph. The answer is I don’t. So I don’t use commas but I don’t miss full stops ever.

    DrLoser there are a limited number of English speaking countries who by their punctuation standard who don’t mandate full stops at end of paragraphs. Do notice that joepeatf early posts have this fault and he is starting to correct to look more general.

    The fact I said commas was if you would in fact use your brain and look closer there is more than 1 difference in punctuation some point to particular nationalities. Then when you look at common word usage it narrows.

    I believe you are free to quote the Macquarie definition here, oiaohm. Have at it.
    There is a reason why I told grece get the book by using the words “Pick up” that most public libraries around the place have it.
    https://www.macquariedictionary.com.au/subscription/login/
    You have to pay subscription to access the on-line form “The Macquarie dictionary”. Yes it is the official dictionary of Australia and is the one used in court cases here.

    By the way DrLoser people quoting “The Macquarie dictionary” defines have been done for copyright infringement. So I am not free to quote the defines here. All I can really do is tell you get the book from a public/school library or pay the online subscription fee or buy the book.

    I do admit I don’t use commas to Australian standard as I don’t use them.

  62. DrLoser wrote, “That’s a very generous nine cores, Robert. Why waste money on the other eleven, long hair or not?”

    Uh, multiple users, multiple processes, stuff like that. There are times when 30 people use the network for instance, during parties, or if I build a kernel I could use even more. Beast is not about throughput alone but service/responsiveness. I like that.

    BTW, the OS uses way more than one core if available.

  63. DrLoser says:

    Anyway, back to you, Robert. This 24-core thing that you will never in a million years stump up $1,250 for. What, in purely general terms, do you think those 24 cores would be useful for?

    Here’s my estimates, based upon a vague knowledge of your typical (multi-tasking!) work-flow:

    1) One core 100% operating system. Could be less, could be that. But you go beyond one core onthe OS, you’re in trouble.
    2) One core handling networking. Depends upon the network connections, but probably about right.
    3) Three cores running databases for things like Word Press, recipes, and whatever else. Probably too many, but hey, best be on the expensive side.
    4) Two cores running your “thin clients.” Again, too many, but laissez les bons temps roulez!
    5) Two cores for all that important Pscal development programming you promise yourself you’ll do, but never get round to. How is that Web GEBC replacement going, btw?

    That’s a very generous nine cores, Robert. Why waste money on the other eleven, long hair or not?

  64. DrLoser says:

    I believe you are free to quote the Macquarie definition here, oiaohm. Have at it.

    Australian English classes the word english as a general noun

    A definition of “general noun,” please. Not a recognised grammatical form, I think.

    … unless its used in a name for a particular individual language.

    OK, fair enough. Not true, but let’s run with it. Not a “general noun” (see your admission above), but it’s a name for a “particular individual language.” Say, “gibberish.” That just about works on your terms. Nobody else’s terms, but then you’re a special snowflake, aren’t you, oiaohm?

    So referring the collective of English languages is english.

    That doesn’t even follow from what passes for your previous arguments, does it, oiaohm, does it?

    Did you grow up stupid, or did your teachers drum stupid into you?

  65. DrLoser says:

    DrLoser moron idiot. Commas was the big one. If you look closer we not even close to the same nationality.

    And what suspected nationality (as if that is at all important when differentiating two gibbering imbeciles, neither of whom can construe a simple sentence in English) would joepeatf be, oiaohm? You’ll notice I am using your preferred handle.

    To repeat, “You forgot to sprinkle in the commas, oiaohm. So, not quite the distinguishing mark you claim.

    You’ve been at the booze again, haven’t you?

    One answer at a time, if you please. One answer is better than no answers at all.

  66. DrLoser says:

    Have you checked the shape of your head, just in case oiaohm wants to (legitimately, not at all offensively) call you a “blackfella?”

  67. DrLoser says:

    Let’s see how long it takes him to remember you from TM Repository, OOP.

    He’s more than capable of finding some stupid and illiterate excuse. “Dr Loser you did not say he differ PR speak. Obvious me all along the similitudiness. This one come from East Asia obvious from the specifics. I knew you did not ask me I tell because honest when *brain haemorrhage*.”

    Go on, oiaohm, guess. Quote the moniker. Back in those days, you somehow believed that Out of phase transistor and I were identical twins.

    Twit.

  68. An Out Of Phase Transistor says:

    You’re giving Robert and his pet hamster,”Fifi,” ideas…

    Let’s see how long it takes for the posts about Russian mafia to stat pouring in.

  69. DrLoser says:

    Actually, thinking about it, oiaohm may simply be in the terminal throes of a drug and alcohol addiction. There’s an eery similarity between his “the Microsoft PR firmm is out to get me” and the delusions of my friend in the cafe down the road — who is, indeed, in the terminal throes of a drug and alcohol addiction.

    Daz (owner of the cafe) has, in all seriousness, suggested that the “unexplained cash losses” at his establishment are down to a conspiracy between his staff and the Russian mafia, who have cloned his till and are clandestinely drawing money from his business account. We’vee tried to point out to him that (a) the Russian mafia has better things to do (b) there isn’t any money in his business acount anyway (c) it’s fairly obvious where the “unexplained cash losses” are going, but he seems quite persistent in his belief.

    At the moment, I’m more inclined to believe Daz than I am oiaohm.

  70. DrLoser says:

    DrLoser really lack of typos/errors is a side effect of being behind one of the masking agents designed the microsoft PR firm.

    You really are a joy, oiaohm.

    Not only do you straw-man a complete irrelevancy (“lack of typos/errors”) that exists only in your head.

    You also seem to have convinced yourself that a “Microsoft PR firm” conspiracy exists. Presumably aimed at you, personally. I suppose it’s easier to believe in this ridiculous conspiracy than to admit that there are at least four — actually more, including Grece and OOP Transistor — people out there who consider your views worthless, yet hilariously silly.

    Incidentally, the “Microsoft PR firm” would be wasting its time with such a wonderful asset. Something that perfect would be worth hundreds of millions of dollars on the open market.

  71. oiaohm says:

    Grece
    Well, first of all, the word English is capitalized
    Pick up Macquarie Dictionary the standard Australian school dictionary look up the word english. Notice it defined with capital and lower case. For how I used it lower case is right. The rest of grece response was just uninformed rant.

    The capitalization of the word english is in the usually camp. Australian English classes the word english as a general noun unless its used in a name for a particular individual language. So referring the collective of English languages is english. Yes it a very warped plural.

  72. Grece says:

    You have very limited understand of the different nationalities of english writing styles particularly in word usage.

    Well, first of all, the word English is capitalized, second your spelling and grammar is beyond terrible. Third,tThe spelling errors, non-parallel sentence structure, and punctuation errors detracted from whatever point the you were trying to make.

    No wonder no one takes you serious.

  73. oiaohm says:

    DrLoser the original fifi is not Australian either. You have very limited understand of the different nationalities of english writing styles particularly in word usage. Everyone who is not a nationality that DrLoser knows the idiot thinks is the same person if they have similar points of view..

  74. oiaohm says:

    DrLoser really lack of typos/errors is a side effect of being behind one of the masking agents designed the microsoft PR firm.
    English as a second language. US/CA, UK. Those are all documented masks.

    Also the PR firm documents writing a fiction to have different attitudes while being different people..

    The only thing that seems to distinguish Fifi Mk I and Fifi Mk II is that, now you have been found out, you throw the occasional (very occasional) hissy fit at each other, you attempt to remember which one is not supposed to be able to use commas, and one of you makes the charming confession that the other one once accused him of “racism.”
    DrLoser moron idiot. Commas was the big one. If you look closer we not even close to the same nationality.

    hence<< When have you ever seen me use that word. Australians don't use that word its not covered in Australian school english. Like it or not DrLoser I am Australian. That other one is not as he use words Australians don't use.

    I pointed out the commas because if you had half a brain left you might have looked closer at words as noticed the nationality miss match.

  75. DrLoser says:

    Deaf Spy Google’s Go is more powerful then Async-await.

    No it isn’t.

    Goroutine use green thread

    No it doesn’t.

    … hence it need not very expensive OS thread context switch.

    Except when it does.

  76. DrLoser says:

    [Drivels pathetically about some weird Microsoft trolling conspiracy] … So no you 4 don’t have identifiable different writing styles when someone uses computer generation masking and I knew that the complete time.

    Deaf Spy is identifiably somebody (highly intelligent) who uses English as a second language.

    The Wiz is identifiably an American (arguably Canadian, but whatever).

    I am identifiably born and raised in the UK — spot the spellings, for example — and Kurks has gone so far as to describe me (not entirely unjustifiably) as a “drunken uncle.”

    Joe Monco is, or was, not quite so easy to pin down. But I’d go with an Australian living and working in the UK.

    Those are four quite separate and distinct writing styles, not to mention attitudes, Fifi.

    The only thing that seems to distinguish Fifi Mk I and Fifi Mk II is that, now you have been found out, you throw the occasional (very occasional) hissy fit at each other, you attempt to remember which one is not supposed to be able to use commas, and one of you makes the charming confession that the other one once accused him of “racism.”

    Don’t come the raw prawn with us, Fifi.

  77. DrLoser says:

    You forgot to sprinkle in commas, Fifi.

    I live to serve.

  78. DrLoser says:

    4MB is the relevant quantity. It’s L3 so can serve any of 24 CPUs and will hold the innermost loops of many processes.

    L3 is just accelerated RAM, Robert. I believe it also matters whether you have a dual-bus system or not … then again, I don’t follow these things.

    I’m not sure what deranged version of computer science you live in, but I find it impossible to envision an “innermost loop” for anything other than a massive matrix calculation that needs to be evicted by the scheduler. Sure, it will happen occasionally, but not often enough to matter. What you’re looking at here is an abject software failure. Extra CPUs aren’t even a band-aid for that.

    Do the math again, Robert. 4MB / 24 cores is ~128 KB per core. At ~128KB, you are going to see evictions. And that’s not even taking into account this magical 256 processes that you run at 100% usage, because if I did, it would be ~16KB per process.

    Not to mention the locking overhead. Or the I/O overhead. Or the disk overhead.

    It’s described as a “development board” for good reason, Robert. It’s pretty darned fine if you want to play with the sort of workload that can “trivially” (as we say in the multi-processor business) parallelise to extra cores.

    Unfortunately, Robert, your workload isn’t anything like that. Nothing like it at all. You’d just be wasting 16 to 20 of those cores.

  79. DrLoser wrote, “you can only keep 32K in L1 and maybe 256K in L2”.

    Check out the specs on the chip… “Cortex-A53 MPCore 24cores, 1GHz, L1 I/D=32KB/32KB,
    L2 =256KB, L3 =4MB”
    .

    4MB is the relevant quantity. It’s L3 so can serve any of 24 CPUs and will hold the innermost loops of many processes. Each core of A-53 has its own L1 and L2 caches but the chip has the L3 cache, a buffer between main memory and the chip. That’s where the busiest processes will hang out.

    The whole system will have 24X256Kb L2 + 4MB L3, a total of 10MB cache. I’d be happier if it were larger but it’s good enough and larger would likely waste more energy.

  80. DrLoser wrote, “the equation is 1GB RAM 1 core”.

    So, 16gB and 24 cores is close to that equation. We have no argument. 80% of 24 is 19… I could run 32gB but I think 16 is enough considering that TLW, I and my whole house run on 4gB+2gB+1.5gB = 7.5gB and what’s in the smartphones. 16gB should last us a while. I’d be happy if my Beast replacement had just 16 cores. This post is about something that’s close and on the market. I could easily afford it and 16gB RAM when I do the transactions for Solo and solar generation. I’m sure it would last me a long while. I did the maths and over 10 years would likely pay for itself in energy savings considering that Manitoba Hydro is raising rates. I could bring in some solar power but that’s not the plan at the moment.

  81. joepeatf says:

    It is not my fault the mentioned products are actually good and working. A 1st generation Surface Pro has still much more value than the latest Chromebook Google produced. Fact.
    Deaf Spy please I suggest alternate products not Linux however not microsoft

    mrpogson DOT com/2018/03/28/going-backwards/#comment-391570
    Microsoft’s c# Async-await is an amazing, precious concept.

    Deaf Spy Google’s Go is more powerful then Async-await. Goroutine use green thread hence it need not very expensive OS thread context switch. I believe you like Go
    I hate Scala very complex programmers language however its not microsoft. I believe you like Scala

    mrpogson DOT com/2018/04/24/on-my-dance-card/#comment-392338
    Btw, I bought my wife a refurbished Microsoft Surface Pro for 250 GBP – i5, 4RAM, 128 GB SSD, with a touch cover keyboard.
    Microsoft Surface Pro is an amazing little device, Robert.

    Deaf Spy please use Macbook Pro. Macbook Pro. is UNIX compliant POSIX also however it is not Linux Deaf Spy hates Linux hence Macbook Pro is alternative

    mrpogson DOT com/2018/04/24/on-my-dance-card/#comment-392301
    Microsoft Visual Studio is mapping a rather huge workspace from TFS locally, plus a backup is restored in a local instance of SQL Server.

    visual studio is bad IDE, it force you use MSBuild. its ok however visual studio force you use generate xml file. Deaf Spy please make C project…please show me generate vcxproj file
    Fact visual studio is bad SCM

    Deaf Spy use Eclipse. Eclipse has custimise builder it use Maven Gradle Ant +a lot

    mrpogson DOT com/2018/04/24/on-my-dance-card/#comment-392226
    Microsoft Windows Server 2016 OS with full-bore asynchronous I/O and fibres

    Linux specialty is network I/O however I not understand why does Deaf Spy hate Linux? However FreeBSD is alternate. Netflix use FreeBSD I believe

  82. oiaohm says:

    Grece
    https://ark.intel.com/products/36500/Intel-Core2-Duo-Processor-E7400-3M-Cache-2_80-GHz-1066-MHz-FSB
    That is the cpu in the computer I am currently using the motherboard P5KPL-AM/PS. I built this machine and its almost up to 10 years old now. This machine was built in 2009. Graphics card had to be updated because it was trouble at about 2013 its now GeForce GT 630 the prior one was being a problem because it require legacy drivers. My next build year for motherboard cpu and ram is 2019. Most likely will push video card until it runs out of driver support for current opengl versions.

    This machine is my internet machine not my build machine. My current build machine will come my internet machine at about the 5 year mark.

    So I have been averaging 9 to 10 years out cpu, motherboards, cases and ram. before I scrap them. I have had to replace hard-drives more often.

    If I have to go second hand is always from proper referibished .

    But you just said that you buy all your hardware new and run it for ten years. Whats the matter can’t keep the lies straight anymore?
    Not at all buying my own hardware was the first line. This is because you don’t want to believe I work for people. I am not always buying hardware for me.

    If my boss is demanding I acquire business second hand its always proper refurbished. Again this is not my choice. My choice is new so I have full history on what has been done to the hardware.

    I have all the invoices for the parts my personal computers are built from and they were all new parts and I assembled them all myself.

    In fact, your typical consumers only keep computers for 3-4 years then they get a new one.
    Go to a Linux conference take close note of the laptops in the room. The average laptop age is 7 years old at a Linux conference sometimes this is insanely clear when you are seen stickers from conferences 8-9 years ago on the laptops. Linux users are not your typical consumers. You basically need to double the typical and then add some. The problem is at 5 years old if you sell the machine of second hand you get bugger all for it. If you want a second machine and you are a Linux user might well use it for another 5 more years.

    Reality is if more people came Linux users PC sales would drop even more due to the longer cycles Linux users keep their machines.

  83. Deaf Spy says:

    Deaf Spy advertise a lot of microsoft product

    Let’s accept this, despite that point 2 is just stating an obvious fact, and the last point is not mine, though I’m flattered to have it attributed to me.

    It is not my fault the mentioned products are actually good and working. A 1st generation Surface Pro has still much more value than the latest Chromebook Google produced. Fact.

  84. joepeatf says:

    WordPress please do not spam filter me

    There was a information leak about how Microsoft trolling company masked their writing style. All 4 of you were inside that profile. So I challenged you to find out. But all 4 of your were already mixing my up with handles that were not mine so mixing you up with each other was just doing what you can done to me. So I thought it was perfectly fair to mix you guys up because that exactly what you were doing to me. Yet you still keep on doing it your moron DrLoser.
    i know right! Deaf Spy is suspicous. Always I have arguement against Deaf Spy, however it makes Deaf Spy advertise a lot of microsoft product. Literally Deaf Spy advertise microsoft like product placement

    mrpogson DOT com/2018/03/28/going-backwards/#comment-391570

    Microsoft’s c# Async-await is an amazing, precious concept.

    mrpogson DOT com/2018/04/24/on-my-dance-card/#comment-392354

    Microsoft Surface models got released some months ago and Surface is scoring an enviable growth.

    mrpogson DOT com/2018/04/24/on-my-dance-card/#comment-392338

    Btw, I bought my wife a refurbished Microsoft Surface Pro for 250 GBP – i5, 4RAM, 128 GB SSD, with a touch cover keyboard.

    Microsoft Surface Pro is an amazing little device, Robert.

    mrpogson DOT com/2018/04/24/on-my-dance-card/#comment-392301

    Microsoft Visual Studio is mapping a rather huge workspace from TFS locally, plus a backup is restored in a local instance of SQL Server.

    mrpogson DOT com/2018/04/24/on-my-dance-card/#comment-392226

    Microsoft Windows Server 2016 OS with full-bore asynchronous I/O and fibres

  85. oiaohm says:

    3) Having “more cores” is only relevant if you can max out at least 50% of them. A typical server at Bing, for example, runs on eight cores with roughly a 80% standard load. I am given to understand that a typical server at Google runs on four to six cores, because they have a slightly different philosophy: buy in bulk and manage the load in dev-ops. But the standard load is still around 80%.
    DrLoser above you have a problem. Google is x86 in there server farms. I guess bing is x86 as well.

    More cores maths of 50% used is due to intel/AMD x86 power management.

    https://www.linaro.org/blog/power-management-of-an-arm-system/
    Power management on arm is very different you are not going into sleep states. Arm in order core in idle is like intel cores in s3 or deeper. So arm in order does not need the same amount of activity to keep cores out of sleep states because they don’t need to switch in and out of sleep states to end up low power using. Instead arm in order normally goes with turning complete blocks of cores off because the deepest sleep state of cut power is about the only one of advantage over idle due to the arm in order core design.

    What was found in arm big little experiments was all cores on worked out quite well. Why most current made arm big little are Heterogeneous multi-processing. Even at 5% load having all cores on in big little showed reduction in processing latency while barely altering the power consumed compared to running all task on the big cores. As long as a core has a thread to process it being on with arm is beneficial because stalled state does not cost badly in power on arm in order cores.

    The difference in power performance with arm in order is why its not as big of problem to expand number of cores. Expanding the number of cores in a arm in order core use chip only minor effects over power effectiveness under light loads. Expanding number of cores on x86 adversely effects power usage under light loads.

    4 core x86 idle use 1/4 of a 12 core x86 idle at best. . 4 core in order arm vs 12 core in order arm idle is the difference is less than 1 percent. Having cores doing nothing in a arm in order system is as not power harmful. Having cores doing nothing with x86 is power harmful.

    Result is with arm systems you will see large number of cores thrown at problem because this is not going to have as bad negitive effect on power usage and results in lower latency.

    There are reasons why you are seeing these massive packs of a53 what is a little core. In order cpu designs allow automatic sleeping as soon as no thread is assigned to core and automatic sleeping as soon as the core is waiting results of a read/write. Out of order processing does bring some serous complexities to power management.

    Yes you don’t want to stall the out of order arm cores because they will also start eating into your power like the x86 cores do as they do need to be sent into sleep states to stop the out of order processing.

    Out of order is a truly a double sided sword. Out of order helps when you have enough IO and are wanting to process fast. In order helps under light lights and IO starvation.

    DrLoser basically there is one lot of maths for in-order processes and different lot of math for out of order processes on how many cores you should point at a problem. In-order is the one with the larger number of cores. This is partly to cover up the performance loss by lack of out of order with the in-order but its also that the in-order power management is done way differently resulting increased number of cores not being as harmful.

  86. joepeatf says:

    There was a information leak about how Microsoft trolling company masked their writing style. All 4 of you were inside that profile. So I challenged you to find out. But all 4 of your were already mixing my up with handles that were not mine so mixing you up with each other was just doing what you can done to me. So I thought it was perfectly fair to mix you guys up because that exactly what you were doing to me. Yet you still keep on doing it your moron DrLoser.
    i know right! Deaf Spy is suspicous. Always I have arguement against Deaf Spy, however it makes Deaf Spy advertise a lot of microsoft product. Literally Deaf Spy advertise microsoft like product placement

    http://mrpogson.com/2018/03/28/going-backwards/#comment-391570

    Microsoft’s c# Async-await is an amazing, precious concept.

    http://mrpogson.com/2018/04/24/on-my-dance-card/#comment-392354

    Microsoft Surface models got released some months ago and Surface is scoring an enviable growth.

    http://mrpogson.com/2018/04/24/on-my-dance-card/#comment-392338

    Btw, I bought my wife a refurbished Microsoft Surface Pro for 250 GBP – i5, 4RAM, 128 GB SSD, with a touch cover keyboard.

    Microsoft Surface Pro is an amazing little device, Robert.

    http://mrpogson.com/2018/04/24/on-my-dance-card/#comment-392301

    Microsoft Visual Studio is mapping a rather huge workspace from TFS locally, plus a backup is restored in a local instance of SQL Server.

    http://mrpogson.com/2018/04/24/on-my-dance-card/#comment-392226

    Microsoft Windows Server 2016 OS with full-bore asynchronous I/O and fibres

  87. Grece says:

    In fact I buy all my own hardware new. Ok I run it for over 10 years from new.

    You don’t buy any hardware, nor do you run it for ten-years. You dumpster dive after your late night sessions. In fact, your typical consumers only keep computers for 3-4 years then they get a new one. Some of it is due to planned obsolescence, by either hardware or software.

    If what Petey says is allegedly true, he is now using a computer from 2008. Hey Petye, post some information about what computer you are presently using. I am using an Inspiron-3847 with an i5 processor, the bane of Pogson’s existence.

    If I have to go second hand is always from proper referibished .

    But you just said that you buy all your hardware new and run it for ten years. Whats the matter can’t keep the lies straight anymore?

  88. joepeatf says:

    This kind of explains why you see these 20+ cores of a53 being in order chips and they perform really well under workloads that you would expect to have IO starvation problems. In order in IO starvation are not kick themselves where it hurts by not doing IO on branches that never need to be executed.
    Have a lot of cores CPU is bad… Server is IO bound always hence CPU is wait a lot.

    oiaohm network+hard drive=service. Server need not need best hard drive however database need 10000rpm a lot of cache also.

    a lot of of CPU core? you need GPGPU not CPU. GPGPU obsolete 20 core CPU, Nvidia has a lot of CUDA core, AMD has a lot also. Reality is CPU is never efficient. All compute bound is to GPGPU.

    Also FPGA is good however it have not floating point ASIC.

    Googles TPU is floating point ASIC. TPU=Tensorflow processer unit. It use machine learning

  89. oiaohm says:

    DrLoser o yes you have claimed you could on TM repository don’t try to lie your way out of it now.

    Welcome back, Doldy. It’s been a while.
    Besides this crap showed you thought you could. Reality accept you cannot DrLoser. In future respond to people using their correct handles or direct short forms from their handles this way you will not screw up.

    Really DrLoser I have never told people to buy anything used on ebay. In fact I buy all my own hardware new. Ok I run it for over 10 years from new. If I have to go second hand is always from proper referibished . There is a reason for this I have been in a business that was bitten buying second hard on ebay that the items turned out to be stolen..

    Although, interestingly enough, you have managed to confuse me, Joe Monco, Deaf Spy and Wiz on multiple occasions, despite the fact that we all have identifiably different writing styles.
    There was a information leak about how Microsoft trolling company masked their writing style. All 4 of you were inside that profile. So I challenged you to find out. But all 4 of your were already mixing my up with handles that were not mine so mixing you up with each other was just doing what you can done to me. So I thought it was perfectly fair to mix you guys up because that exactly what you were doing to me. Yet you still keep on doing it your moron DrLoser.

    So no you 4 don’t have identifiable different writing styles when someone uses computer generation masking and I knew that the complete time.

    But how often to I not call you by your handle.

    Reality is fifi does not have anything to-do with me either. Its another person handle.

    Big thing here those people recommend doing things that I never recommend DrLoser.

  90. Dr Loser says:

    As a general rule of thumb, btw, there is actually a desirable relationship between the number of cores and the amount of RAM. (Cache doesn’t count, for tolerably obvious reasons.)

    Roughly, and forgetting for the moment that you need one core and substantial RAM to deal with the OS kernel alone, the equation is 1GB RAM 1 core. The actual clock speed of the core isn’t really relevant here, either, because we’re talking about process space and avoiding swaps to hard drives.

    Do the math for your silly little yet extraordinarily expensive 20 core motherboard, Robert. Take your socks off if necessary.

  91. Dr Loser says:

    Keeping stuff in RAM and cache and having more cores works for me.

    As Deaf Spy points out (as if to an idiot red-headed step-child), Robert, you don’t know because you’ve never tried it. But let’s deconstruct that fatuous little assertion, shall we?

    1) Keeping things in cache. Well, fer shure, me old Manitoban mucker. But the thing is, you can only keep 32K in L1 and maybe 256K in L2. I think you’re going to have a very hard time running “256 processes” — no matter whether they are in run- or wait-state — in that amount of cache.
    2) Keeping things in RAM. Here you are on safer ground. However, the amount of RAM has little or no bearing on cross-core efficiency. It’s merely a matter of avoiding swapping to disk.
    3) Having “more cores” is only relevant if you can max out at least 50% of them. A typical server at Bing, for example, runs on eight cores with roughly a 80% standard load. I am given to understand that a typical server at Google runs on four to six cores, because they have a slightly different philosophy: buy in bulk and manage the load in dev-ops. But the standard load is still around 80%.

    Both approaches work. Neither really involves RAM or cache. And if you were to tell any large server provider that they could really do with 20 cores running at 1 GHz … it doesn’t actually matter which architecture you propose. They wouldn’t even waste the breath to laugh in your face.

  92. Dr Loser says:

    This is so funny for years you have claimed you could identify me no matter if I changed my handle.

    I’ve never claimed that once, Fifi.

    Although, interestingly enough, you have managed to confuse me, Joe Monco, Deaf Spy and Wiz on multiple occasions, despite the fact that we all have identifiably different writing styles.

    As opposed to, say, Fifi Mk I and Fifi Mk II.

  93. oiaohm says:

    Drloser take a closer look at my writing under oiaohm. There are words I never shorten as well. The grammar you are using to so called find me is a grammar that finding multi people and always has. Its because you are a incompetent moron.

    Really Robert please point out to Drloser the super moron that joepeatf and me are not the same person and I was just away for 3 weeks. Maybe then DrLoser will give up on his stupidity of call me names that are not me. And maybe then the other idiots will also stop it.

  94. oiaohm says:

    First, you think HT is threads spawned by the CPU. No, Fifi, this is not what HT is all about.

    Deaf Spy I see problem. I said starts. CPU design a CPU starting a thread is not a thread being spawned/created. When you are talking about HT. You end up talking about starting and stopping threads with state save and restore. So this means you don’t understand HT at a silicon design level. So I have been talking over you skill level.

    joepeatf don’t worry Deaf Spy and DrLoser are total morons who don’t want to believe that more than 1 person can disagree with their point of view. Worst cannot even use a person correct handle.

    Sorry, oiaohm, you’re going to have to try an awful lot harder than that. It’s pretty clear that you swapped schizo-personalities for three weeks or so, and that there was no overlap at all, and that you are now making a pathetic little effort to pretend that you are not the same poster.

    This is so funny for years you have claimed you could identify me no matter if I changed my handle. And you are failing completely. Drloser thank you for the best laugh ever.

  95. Deaf Spy says:

    Four cores of A-53 at 2gHz is about the same power as 24 cores at 1gHz.

    Robert, you will have a very hard time to provide this. Very hard. Because it is utterly wrong. You may find it performing the same in a certain very specific scenario, but I greatly doubt you can find or event construct one.

    I have way more than 24 processes running so I can keep way more than 4 cores useful

    Irrelevant, because most of your processes are waiting, not running.
    What does a web server do when there are zero incoming requests?

    I don’t care much about the clockspeed just that when I click something happens soon.

    Nice, but define soon. If your click involves an IO and your IO is slow, no amount of cores will help you out.

    Keeping stuff in RAM and cache and having more cores works for me.

    You don’t know that, Robert. You’ve never tried it.

  96. DrLoser says:

    But, back to Manitoba:

    More cores reduces power-consumption for a chip.

    No it doesn’t.

    For half the nm one gets four times the cores

    … but (if you are lucky) precisely the same power consumption. Do the math, Robert. Modern chip fabs half the nm and half the clock (roughly speaking: Intel went from ~3.8 GHz to ~1.8 GHz even as they went from 22nm to 12 nm) precisely in order to get “four cores for one” with the same die size. Interestingly enough, the requirement for heat dissipation is greatly reduced, solely because the design is simpler. Check out the current size of heat sinks for Intel chips, as compared to the monster heat sinks of ~2008.

    If you hold everything else constant, and you just increase the number of cores from 1 to 4, then you’re still dissipating the same amount of heat. And that assumes that 12nm tech generates the same amount of heat as 22nm heat. Which it doesn’t. It’s hotter..

    and by reducing the clockspeed a bit a reduction in TDP as well.

    One would assume that there is something like a linear relationship between clock speed and TDP, Robert. (Right up until the point where Moore’s Law ceases to operate, which is a limit you have yet to acknowledge ) This is not news, nor does it have anything to do with the number of cores, nor does it have anything to do with the number of cores you wish to pack in, nor does it have anything to do with, and this is important, the ability to use those extra cores.

    Do try not to sound like your bloviating imbecile friend Fifi.

  97. DrLoser says:

    “I/O starvation?” Boggle boggle boggle.

    Imagine a single core machine running a LAMP stack. “I/O starvation” would basically imply that you’re waiting on a request and you don’t get one for the next, say, 1/10th of a second.

    Result? 10% CPU usage at worst.

    Now imagine the same “I/O starvation” on a 20-core ARM machine. Seems a bit of a waste of 19.8 cors to me, Fifi.

    Boggle boggle boggle.

  98. DrLoser says:

    There is a jitter factor is something you have not considered DeafSpy.

    Boggle boggle boggle.

  99. DrLoser says:

    And now we come to the slightly (but only slightly) less idiotic version of Fifi.

    Really out of order cores when you get to IO starvation can make maters worse than if you are using a stack of inorder cores.

    Boggle. Boggle. Boggle.

  100. DrLoser says:

    Captain obvious microsoft made windows 10, Bob buys laptop+windows 10 preinstalled.

    I think you’ll find that “Bob” has done no such thing. I’m open to being corrected by “Bob.”

    Apparently Peaty has decided to be the even more idiotic of the Fifi twins.

  101. DrLoser says:

    Deafspy name calls me fifi also! Robert Pogson does not name calls me fifi , however everybody else harrass me.

    Congratulations, Fifi, you managed to insert a single comma in there. What a diference that makes, in Fifi-Land!.

    Unfortunately, that entire sentence is basically Fifi-grammar.

    Sorry, oiaohm, you’re going to have to try an awful lot harder than that. It’s pretty clear that you swapped schizo-personalities for three weeks or so, and that there was no overlap at all, and that you are now making a pathetic little effort to pretend that you are not the same poster.

    It’s fun to read, though. Keep trying! Perhaps one of you will learn English along the way!

  102. Deaf Spy says:

    Fifi and Fifi, you are both having bad day.

    First, you think HT is threads spawned by the CPU. No, Fifi, this is not what HT is all about.

    Then, you say that MS are killing Surface, when new Surface models got released some months ago and Surface is scoring an enviable growth.

    Did you take your meds today, sweethearts?

  103. joepeatf says:

    Deafspy you must be blind. There is no Fifi here. So stop demarding a person who not answer something. Don’t bother asking me either because I am not fifi and I am not going answer questions you have directly asked to fifi.
    Deafspy name calls me fifi also! Robert Pogson does not name calls me fifi , however everybody else harrass me.

  104. joepeatf says:

    In the other news, Robert, Microsoft reported:
    “Windows OEM revenue increased 4% (up 4% in constant currency) driven by OEM Pro revenue growth of 11%”
    followed by:
    “Surface revenue increased 32% (up 27% in constant currency) against a prior year comparable impacted by product end-of-life-cycle dynamics”

    Captain obvious microsoft made windows 10, Bob buys laptop+windows 10 preinstalled. Also microsoft surface is dieing,Microsoft Surface Tablet Usage Market Share Is Just 1.6% In The US https://mspoweruser.com/microsoft-surface-tablet-usage-market-share-is-just-1-6-in-the-us

    recent microsoff killed windows phone, hence windows surface dies also,Microsoft finally admits Windows Phone is dead https://www.theverge.com/2017/10/9/16446280/microsoft-finally-admits-windows-phone-is-dead

    You buy a lot of microsoft product Deaf Spy, however I believe you should avoid microsoft product. microsoft has a lot of EEE embrace extend extinguish. They kill windows phone, They kill windows surface.

    I believe Deaf Spy buyed a windows phone. windows phone is dead, Im not afraid.

  105. joepeatf says:

    PCs are going nowhere, Robert, and ARM is not taking over the desktop. Chromebooks were not even a flash in the pan.
    Deaf Spy please use hear aids. I hate microsoft however even microsoft hate x86.
    microsoft will stop x86 and replace to ARM. Apple will stop x86 and replace to ARM. also.
    https://www.zdnet.com/article/windows-10-on-arm-it-will-be-more-limited-and-heres-how-reveals-microsoft/
    long term microsoft kill Intel monopoly. they move to ARM, its less restriction.

    https://mspoweruser.com/windows-10-on-arm-extensively-benchmarked-natively-and-with-x86-emulation/
    microsoft windows ARM makes x86 emulation

    https://www.cultofmac.com/109835/intel-apple-switching-macbooks-to-arm-is-a-very-real-and-scary-threat-for-us/
    I want to choke author, however Apple will use ARM long term. Rosetta use Powerpc emulation in x86. Now Rosetta2 use x86 emulation in ARM.

  106. oiaohm says:

    Really out of order cores when you get to IO starvation can make maters worse than if you are using a stack of inorder cores.

    In order you get to a point where it cannot perform operation it stalls until either operating system scheduler gives it something else or cpu management system gives it something else.

    Out of order is great for when you are having spare IO and are attempting to process more in a short time. Out of order not that great when you get to a IO starvation problem because it normally makes the starvation worse.

    Out of order can attempt to pre execute thus be requesting more IO instead of stalling and waiting for ability to perform IO to come.. Remember with out of order branch prediction it may be running down branches of code that never need to be executed and will not be executed on a in order core so out of order is performing not required IO to run the program.

    This kind of explains why you see these 20+ cores of a53 being in order chips and they perform really well under workloads that you would expect to have IO starvation problems. In order in IO starvation are not kick themselves where it hurts by not doing IO on branches that never need to be executed.

    Remember your normal intel x86 are out of order processors. So a 4 core intel with 8 thread hyperthreading can be quite a bit slower than a 4 core no hyper-threading in order in IO starvation at the same clock-speed. Why because the hyper-threading and out of order starts is also consuming the IO.

    Cpu as a whole may change what thread is on what core by itself. This can also come from the power management cpu inside most modern cpu or thread management engine in some multi core chips. Why if one particular area of the chip is get too hot threads can be transferred to other cores and OS and user will never know or you might see it as performance jitter as the power management system inside the cpu tells the core to stall while temperature drops. Hopefully power management stalled core is not something holding a lock all the other threads on all the other cpu are waiting on because this can effectively stall the complete system. The fact of the modern day chip non predictability you don’t want OS wide locks and you want as little OS locking as you can..

    IO issues is why you see the risc-v labelled prototype this does have a IO management engine inside the cpu core that does intentionally stall cores based on priority of the IO operation so lower less critical IO gets delayed. Interesting enough control IO can result in higher throughput due to allowing locking and unlocking of resources to function in more timely fashion as you get closer and closer to true 100 percent IO load. Its gets really interesting when you have SMT/thread management and IO management inside a cpu.

    Yes intel x86 chips have been around for a long time but this does not mean they have gone down all the right paths in design. Some of the nightmare here is some choices are intel has made is only beneficial under particular conditions and under other conditions those things are a hindrance.

    You cannot design a CPU core that is universally good at everything. Some of the reason why is a lot of consideration going into mixed cpu types in a single chip. So you can have in order and out of order cores and allocate tasks where suitable. If you are in a heavy IO bind you can be wanting to have everything on in order cores.

  107. oiaohm says:

    So cpu starts a thread this consumes time find out it cannot perform IO then has to switch to another thread that may be able to perform IO.

    More utter non-sense. CPU don’t start threads. Programs do. And you have no idea how IO works
    Someone is clueless about hyper-threading and multi core thread management engines.

    A, btw, Fifi, don’t forget to answer – what is the scaling Windows 10 defaults to when connected to a 4K display?
    Deafspy you must be blind. There is no Fifi here. So stop demarding a person who not answer something. Don’t bother asking me either because I am not fifi and I am not going answer questions you have directly asked to fifi.

  108. Deaf Spy says:

    In the other news, Robert, Microsoft reported:
    “Windows OEM revenue increased 4% (up 4% in constant currency) driven by OEM Pro revenue growth of 11%”
    followed by:
    “Surface revenue increased 32% (up 27% in constant currency) against a prior year comparable impacted by product end-of-life-cycle dynamics”

    PCs are going nowhere, Robert, and ARM is not taking over the desktop. Chromebooks were not even a flash in the pan.

    Btw, I bought my wife a refurbished Surface Pro for 250 GBP – i5, 4RAM, 128 GB SSD, with a touch cover keyboard. This is an amazing little device, Robert. No one with one’s senses will choose any form of a thin client, compared to it. No one.

  109. Deaf Spy says:

    A, btw, Fifi, don’t forget to answer – what is the scaling Windows 10 defaults to when connected to a 4K display?

  110. Deaf Spy says:

    If this was a Linux workload being on a 10 core 1/4 the clock-speed might be performing identically or better.

    Utter non-sense.

    So cpu starts a thread this consumes time find out it cannot perform IO then has to switch to another thread that may be able to perform IO.

    More utter non-sense. CPU don’t start threads. Programs do. And you have no idea how IO works.

    The rest is even worse utter non-sense. The link is, as usual, irrelevant to the discussion.

    Go dig some better links, Fifi.

  111. oiaohm says:

    At the moment I write this, my 2-core 4MB cache laptop is running 220 processes, 2700 threads, doing quite some heavy IO-intensive work – VS is mapping a rather huge workspace from TFS locally, plus a backup is restored in a local instance of SQL Server. The CPU load is… between 50 and 70%. Obviously, more cores won’t achieve a thing. I am clearly limited on IO at this moment.
    Deaf Spy other than it being a windows workload. If this was a Linux workload being on a 10 core 1/4 the clock-speed might be performing identically or better. In fact in some cases faster.

    The power effectiveness of arm cores or risc-v cores means you don’t need to power them off to prevent chip from overheating.

    There is a jitter factor is something you have not considered DeafSpy. The reality is the IO may not be 100 percent filled. Not all threads/processes running will be performing the same IO. So cpu starts a thread this consumes time find out it cannot perform IO then has to switch to another thread that may be able to perform IO.

    This is 2 lanes vs 10 lanes. Making 2 lines process 4 times faster than 10 lanes does not mean they 2 lanes will clear the workload faster. The faster you clear the cpu workload the more the IO can be held a 100 percent with less jitters in the IO speeds.

    Its a trap for the not aware. Of course those from windows don’t allow that Windows internally stuffs up on locking so effectively stalling all cores
    https://randomascii.wordpress.com/2017/07/09/24-core-cpu-and-i-cant-move-my-mouse/

    Windows desktop still has a big kernel lock that is triggered when you perform different IO so stalling all cores of the cpu. So that 50-70% cpu usage might have nothing to-do with IO might simply be windows desktop tripping over its own design of it locking. Also Windows reports IO is 100 percent full when its not when it tripping over particular locks as well.

    Reason why big kernel lock removal from Linux kernel was big news because it was the start of the process to stop Linux tripping all over it self on multi core systems.

    No, they don’t. Modern chips, when not in use, automatically lower their power consumption.
    Deaf Spy they don’t do it automatically the kernel driver of the operating system has to tell power management process for intel(this is ME) that then tells the CPU to go into sleep modes. The Linux kernel had a issue recently of telling ME to put CPUs to go into a sleep mode that worked on older intel multi core cpus to save power. But on some of the modern intel cpus you in fact save power by not powering the cpu down at all into that mode. Yes a sleep mode that eats more power than the cpu running 100 percent load and worse generating twice as much heat as max load. So it is fairly important to make sure your operating system power management configuration is right. Reality is power management is a game of hardware quirks something that worked in 1 generation of intel hardware to save power in the next generation can be the worse power hog ever.

    You hope the OS effectively lowers power consumption when the core are not in use.

    An Out Of Phase Transistor suggesting a massive multi core system running Windows 7 means he is clueless. Reality is that you would put Windows server or Linux on a multi core system if you can or you will have windows desktop tripping over its own feet killing performance. Of course this appears like its filled the IO when in reality the cpu can be doing nothing and the IO can be nothing for a lot of the time.

  112. An Out Of Phase Transistor says:

    Well no, AMD had lower than Intel frequencies, because they could get away with it, as Intel ran into a wall with the P4 design in late 2003, then failed to adapt quickly enough.

    Back then, some tried to run Pentium M (single core, no HT) chips on desktop motherboards, it worked BTW, those too had very low clockspeed but were perfectly competitive with the P4 series.

    It’s strange how long Intel took to ship a Pentium M derived series of processors as their desktop line (in an official and supported capacity, not as third party hacks), but once they did, they a) immediately retook the market (not that they lost much of it in the first place) and b) scaled clockspeeds upwards toward 4GHz (enthusiasts went higher, apparently without much trouble).

    The link between multi-core and low clockspeeds exists only as a splinter in your eyes Robert.

  113. DrLoser, being really stupid, wrote, ” Do the maths.”

    The whole world followed AMD down the 64-bit rabbit-hole for a reason. More cores reduces power-consumption for a chip. For half the nm one gets four times the cores and by reducing the clockspeed a bit a reduction in TDP as well. Keeping the clockspeed fixed would increase TDP. This is old news. Four cores of A-53 at 2gHz is about the same power as 24 cores at 1gHz. It’s not a simple relation but it works for me. I have way more than 24 processes running so I can keep way more than 4 cores useful and I don’t care much about the clockspeed just that when I click something happens soon. Keeping stuff in RAM and cache and having more cores works for me. It seems AMD, Intel, ARM and the rest of the world understands this but not DrLoser.

  114. An Out Of Phase Transistor says:

    Turning off the chips, good god; with a hammer, right, Robert?

  115. An Out Of Phase Transistor says:

    Oh, sorry, I, too wondered whether Robert simply wanted the Wizzard to break out laughing so hard as to hit the screen then fall over backwards… that is to say: was it an attempted assault, thinly veiled as joke.

    Or was it an unintended joke, poor as it may be, preposterously offered as an actually meaningful response; the distinction lies, I suppose, in the verdict: aggravated assault vs. assault by aggravation.

  116. DrLoser says:

    If one does the maths, more and smaller cores does reduce the power-consumption for the whole system.

    Fair enough, Robert. Do the maths.

    In front of us.

  117. DrLoser says:

    I could make the same argument for tires on a car. Where’s the need for four of them except to haul around the bloated weight of these gargantuan systems have four empty seats, an empty trunk and a 150HP engine running at 800 rpm in overdrive?

    A very good point about tires, Robert. Four tires good, two tires better.

    But if you really want to save money on tires, you should aim for a unicycle.

    You see where this is going? Cost/Benefit, you feeble old fool. The minimal cost of those extra three tires is dwarfed by the benefit of having three or four of them.

    If you can’t even grasp that obvious point, I’m not sure why you’re still pushing 24 underpowered cores on a barebones motherboard that costs $1,250 that you will never, ever, pay, not in your wildest dreams, not even if it comes down to a third of that second-hand, and besides, you are on recent record as wanting “brand new.”

    Is there anything at all that you would open your wallet for, Robert? No evidence so far. All you have is farcical excuses which don’t even have any sort of engineering significance.

  118. DrLoser says:

    I apologise to the Wiz for this, but I’m tired of waiting for somebody to call Robert out on his stupendous SyAdmin ignorance.

    Sez the Wiz:

    “System administrators keep the utilization at a reasonable level and turn off the chips not needed at certain times of the day.”
    Interesting. The only time I saw System administrators interacting directly with cores this was with an IBM Power Mainframe, Where they had the ability to add 10th’s of a CPU as needed to an LPAR.

    Sez the Pog:

    The Wiz wrote, “How about a link showing this done in the case of an ARM cpu?”

    Uh, the power button or shutdown -h now? You need a link for that?

    Now, I am fair and impartial and I live to serve, Robert, but in this case you need a reverse lobotomy.

    shutdown -h. which btw is not my preferred *nix shutdown command, is marginally preferable to hitting the power button. I’m actually surprised that you didn’t recommend pulling out the plug.

    Shutting down and halting isn’t really the answer to the rubric, is it? Let me remind you of the rubric:

    The only time I saw System administrators interacting directly with cores this was with an IBM Power Mainframe, Where they had the ability to add 10th’s of a CPU as needed to an LPAR.

    Not something you can do with shutdown -h.

    Then again, unfair rubric. Let’s try your rubric, Pogson:

    System administrators keep the utilization at a reasonable level and turn off the chips not needed at certain times of the day.

    shutdown -h doesn’t really work on that scenario, either, does it?

    Turning off the chips, pah. You really are losing it, old man. You can’t even conceal it any longer.

  119. DrLoser says:

    I love how Robert admits that he has no clue what he’s talking about, and just comes straight out with a personal insult in the first sentence. Much like Fifi, as it happens. Although I am fairly sure that they are two different entities.

    Unimaginative idiot! There is a wait-queue. Processes in those queues run on certain cores. Other processes may run on a core that has a process waiting.

    Leaving aside what Deaf Spy’s “imagination has to do with it, although it’s pretty clear what Robert Pogson’s “imagination” has to do with it (clue: delirious) … I have a technical question to ask Robert here.

    You have no clue how the Linux kernel schedules user threads, obviously. But here’s something more appropriate to a 24-core system.

    Do you understand what “thread affinity” means?

    Oh, and PS, unless I missed something, you have yet to admit that you have ever programmed in a multi-threaded environment.

    It’s like talking to a dustman about global warning, isn’t it? You get the occasional random squawk, but basically nothing very useful at all.

  120. DrLoser says:

    Not mentioned: enslavement to the obsolete Wintel monopoly and obscene power consumption. Used? I prefer new.

    I must say, you give value for money, Robert.

    1) When is the last time you bought a new server?
    2) Why do you insist that Intel and/or Microsoft are “obsolete?” In cash terms, they are still both coining it. Also in tech terms, but you are a neanderthal, so we’ll stick with cash terms.
    3) “Obscene power consumption?” I told you about my Intel NUC about eighteen months ago, Robert. It consumes about 12 Watts. A quad core will consume 16 Watts.

    You have a very unusual definition for “obscene,” Robert.

  121. An Out Of Phase Transistor says:

    Enslaved?

    Care to explain that one?

    Not with “fantasies about MS/Intel/IBM/etc” (which is your usual drivel) but rather by explaining “how are you not enslaved with some alternative you’ve come to fancy.”

    Regarding power; what’s bothering you here, money or resource consumption?

    Spoiler: either way, you’re not getting ahead without govt intervention at the national level, and once they do intervene (which they will, eventually), everything available will become “used,” as it is manufacturing “new stuff” that represents the biggest cost, in real terms.

    Sorry to break it to you Robert, but apart from the fact you base your “plans” on fantasy grounds, your actual demands (some of them would even be sensible if you understood them) lead to one way only – govt sponsored monopoly, with limited personal choice (this reduces costs both financially and materially; and is otherwise known as “socialism”), once you realise that, the next question is “What would serve the needs of the greatest number of people, best.” it all comes down to a mean answer:

    Microsoft. Windows.

    Which is the cheapest and actually best available solution and has been for the longest time.

    (Don’t cry like that Robertina, the truth is you were always licensed, your blog’s mad existence, completely futile…)

  122. Someone wrote, “I went and took a peek at what Euro 1200 buys one in used hardware”.

    Not mentioned: enslavement to the obsolete Wintel monopoly and obscene power consumption. Used? I prefer new.

  123. Deaf Spy says:

    There is a wait-queue.

    No, Robert, there is no such thing as a wait queue. There is a scheduler. Get this through your thick head, because you miss also elementary knowledge in operating systems.

    So, there is a scheduler, which schedules only the threads that are ready to run. All threads that are in waiting state (f.e. no work to do as it may happen in tread pools, or for an IO to complete in a synchronous program, or to get a message from the Windows manager, etc., etc.) will never get any CPU time whatsoever. Provided, of course, the scheduler is a competent one. Like in Windows.

    Lots of processes have some tight little loops running in the innermost loop and they can dodge in and out of caches at a great rate.

    Really? Can you please give an example? Because, I can tell you – they don’t.
    Unless it is a huge computation, where lots of data is involved, and then caches are mostly irrelevant.

    At the moment I write this, my 2-core 4MB cache laptop is running 220 processes, 2700 threads, doing quite some heavy IO-intensive work – VS is mapping a rather huge workspace from TFS locally, plus a backup is restored in a local instance of SQL Server. The CPU load is… between 50 and 70%. Obviously, more cores won’t achieve a thing. I am clearly limited on IO at this moment.

    As I recall “move this to there” is about the most frequent sequence in software. Caches love that. Chips with caches love that.

    Unless that there happens to be a dynamic location – an item from an array, accessed by index, and the array doesn’t quite fit in the cache. And, of course, when data is in RAM to start with.

    Speaking of RAM. If there is not sufficient RAM to store the data for all cores, something terrible happens – hard page faults.

    Even when you have lots of RAM and many cores, another not very nice thing may happen – NUMA.

    Btw, Robert, I asked a fair question. Do you even know how asynchronous IO works?

  124. An Out Of Phase Transistor says:

    I, er… I suddenly had one of those urges again, after reading about Roberts wildest desires…

    So I went and took a peek at what Euro 1200 buys one in used hardware; now, I do apologise for such terribly poor behaviour, after all, it may force Robert to climb up a tree or some such (and then we’ll be needing the fire dept.).

    It’s all Windows and Intel etc… did I say I’m sorry? I’m not… ha, ha, ha.

    And so it goes:

    HP Z620 Workstation:
    – Two E5-2643 Xeons (a total of 8 cores and 16 threads at 3.3GHz)
    (oh, not to forget, it has 10 MB of L2 cache, per CPU)
    – 64 GB of memory (this isn’t “up-to,” it’s the actual amount of RAM you get)
    – 1TB HD (thrown in complimentary, I assume)
    – Quadro K4000 (3GB GDDR5, also complimentary, probably)
    – Windows 7 Professional

    1250 Euro (before VAT), with 24 month warranty and 1 year post warranty servising package.
    (SSD upgrades are of course available, and I know I need to point this out, lest I scare any HD-fearing rodent out there.)

    But, but… could it serve; I mean could it? Or would it try to eat the Beast and ascend as the new ruler of the Pogson household, with the TLW only cooking for the Machine from that day onward?

    Would Robert ever post again? Or would he be banished from (or, maybe, to) the basement by the New Management!?

    It really is all fantasy, when it comes to Bob’s purchase plans, isn’t it?

    Robert? How’s that ARMed plan coming along in view of your wallet? Is it thin? The plan, I mean…

    *(Also, I went and tried to put together something new to see what I could get for the money, but then I remembered how futile it would be to go with something reasonable – notice how the proposition put forward by DrLoser didn’t earn a single mention – so I reneged on such sillines… )

  125. Deaf Spy wrote, “There is no such thing as a wait-queue for each core, Robert. No one does it.”

    Unimaginative idiot! There is a wait-queue. Processes in those queues run on certain cores. Other processes may run on a core that has a process waiting. Get with the programme. These are multi-user/multi-tasking systems. It’s not about a single thread. The more cores you have the more processes and users one can please. There are limits where the bottleneck is to RAM or whatever, but 1, 2, 4 or 24 cores are not generally limited by RAM. Lots of processes have some tight little loops running in the innermost loop and they can dodge in and out of caches at a great rate. As I recall “move this to there” is about the most frequent sequence in software. Caches love that. Chips with caches love that. A 24-core chip will do amazing things with that.

  126. Deaf Spy says:

    These systems have huge numbers of processes and there is a wait-queue for each core.

    There is no such thing as a wait-queue for each core, Robert. No one does it. There are very rare cases when one might want to bind a thread to a particular core, and these are primarily testing purposes.

    System administrators keep the utilization at a reasonable level and turn off the chips not needed at certain times of the day.

    No, they don’t. Modern chips, when not in use, automatically lower their power consumption.

    On a modern system 24 cores can certainly be well-utilized.

    Only if paired with very potent IO (both network and storage). Otherwise, your cores will do nothing but wait for IO to complete.

    Neither must we stop at four cores even for small systems.

    No, we don’t. We simply pair the multi-core systems with adequate hardware. For example I run a six-core CPU at home. But I have also 32GB RAM, and two SSD M.2 drives (> 2.5GB/s seq. reading speed). I purchased the MB carefully. So, when I run a few VMs, they don’t choke on IO or page faults.

    Your board hasn’t the potential to drive these.

    If one does the maths, more and smaller cores does reduce the power-consumption for the whole system. This is a perfectly good compromise against system throughput/speed in many cases like my desktop and server.

    No, they don’t quite do so, actually. Simply because modern, powerful cores:
    1) Burst to full speed and complete the work sooner than a slow core.
    2) Consume very little when idling.

  127. oiaohm says:

    Interesting. The only time I saw System administrators interacting directly with cores this was with an IBM Power Mainframe, Where they had the ability to add 10th’s of a CPU as needed to an LPAR.
    https://wiki.linaro.org/WorkingGroups/PowerManagement/Archives/SchedMc?action=show&redirect=WorkingGroups%2FPowerManagement%2FArchived%2FSchedMC

    wizard emiritus arm is kind of strange. Do notice they reference cpu hotplug.
    https://en.wikichip.org/wiki/socionext/sc2a11#Block_diagram
    Notice that the are 12 cpu blocks containing 2 cpu cores in the diagram. Each turns out to be a virtual cpu socket on arm. Yes when a section goes into unpluged state full power to that area of silicon is cut off.

    The number of sockets on the circuit board and the number of sockets you see from software with armv6/64 bit don’t always line up. From software you can be seeing a lot more.

    So internally massive multi core arm chips are very much like of the old mainframe hardware were l3 was on the motherboard with multi cpu sockets sharing it except now all this is inside 1 soc package.

    A lot of what being done in modern day massive multi core soc chips is old mainframe designs done smaller without the individual physical parts but from software you see what would have been the individual parts due to the management interfaces like hot unplugging and plugging cpu sockets that are just construction inside the soc chip..

  128. oiaohm says:

    Spectre problem I mentioned because this causes a problem when you switch from userspace to kernel space and back this equals having to flush the l1 cache. This has done a number on Intel performance in particular areas.

    Having a bigger and bigger L1 cache does not in fact help you when you work out how often you have to flush it. Having more cores to allow more threads to run uninterrupted equals less flushing.

    https://en.wikichip.org/wiki/socionext/sc2a11#Cache
    https://www.7-cpu.com/cpu/Haswell.html

    DrLoser go read. The sc2a11 L1 cache total is 1.5 MiB. Haswall total is 64kb. If a single core is operating on a Haswall is has as much l1 as a individual core in a sc2a11.

    Instruction set cache in sc2a11 is share between 2 cores in the sc2a11 but in the haswall it is shared 8 ways. Data cache in the sc2a11 is shared between 4 cores where the Haswall is shared between 8.

    Next is the sc2a11 l2 is 3meg where the Haswell is 256 KB.

    Haswell l3 is 8 meg but the sc2a11 is 4meg l3. Both chips have similar amounts of cache but with sc2a11 more is in the higher performance l1 and l2.

    Risc-v tried a l1 larger than 32kb. Turns out 32kb is about as big as we can make a l1 caches that runs at the same clock speed as the cpu core otherwise you get distance latency problems at the 1ghz or faster clockspeeds at this stage. Only way to make l1 large is make it less shared and connect it to more cores..

    Lot of the reasons why a 4/8 core intel will be running on 2 cores is power usage.

    Do note what you said most individual work cases are running on 2 cores. Ok why if most usages are 2 cores why is intel sharing l1 caches 8 ways. sc2a11 l1 cache configuration is closer to real world work cases with instruction cache shared 2 ways and the data cache shared 4. There are reasons why sc2a11 at 1ghz under a lot of work cases match up to x86 chip of higher clock-speed.

    Skylake intel alters the per core caching to be closer to a sc2a11.

    Having 1.8 ghz instead of 1ghz is not enough of a gain for compatible workloads..
    https://en.wikipedia.org/wiki/Amdahl%27s_law#/media/File:AmdahlsLaw.svg

    4ghz skylake 4 core with hyperthreading with compatible work loads in performance matches a 1ghz 24 core sc2a11 but the skylake is eating a heck load more power. For a automated built and QA server where the workloads get to 99.9% parallel there is no question you would choose sc2a11 over 4 core skylake if the costing was not an issue..

    The other thing is with modern intel they have really bad power management when you get up into the 12 core versions. Your desktop intel 4 cores only 2 core on you saving power. You intel 12 core server 1 core on you might as well turn another 3 on.

    There are quite a few major performance vs power problems in the intel chips. In the non intel chips there is a issue that the raw clock speed per core is a little low. If the clock-speed was half that of the intel chips it really would not be a question the arm chips would win in most workloads in that case due to the mass volume of cores. This is why a 1.8 x86 4 core vs 1ghz 24 core arm results in the x86 losing due to not being powerful enough.

    Drloser how many tasks do non games but general user stuff use that really use more than 1ghz of cpu processing power. The sc2a11 would need more cores on where the x86 would have like 2 on you would see sc2a11 having like 8 to 10 on. Even so a sc2a11 with 24 cores on is still eating less power than an intel chip with 2 cores on.

    Attempting to run a heavy single threaded game or rapidly trans-code a single video using a single threaded codec on a sc2a11 is not going to be too great in perfomance. But running firefox with chrome with libreoffice all a the same time is not going to perform badly because the workload can be spreed over the mass volume of core the sc2a11 has to play with.

    Really I would say most general software is not as multi core friendly as it could be yet. Lot of it is not disaster bad either due to a lot of the general software that is single threaded not in fact needing a processor faster than 1ghz as long as it can have that complete core to itself.

    This is a serous question is anything that robert uses in fact need a cpu faster than a 1ghz a53 core can do. Do remember machine robert is looking at does take a video card so you can off load video decode.

    From my personal point of view I would be happy for it to be 1.5 to 2ghz instead of 1ghz.

    My biggest issue is not the performance side but instead its the cost and the limited pci-e. In fact the limited pci-e is more likely to be a bigger problem than the cpu processing power.

    This is, I am afraid, because you are an imbecile.
    No DrLoser you just describe yourself. You prove you an imbecile every time you resort to name calling. Really you called me fifi so that you could say pack of lies and not get called out on it right DrLoser. It seams to be every time you name call someone you are making crap up DrLoser.

  129. The Wiz wrote, “How about a link showing this done in the case of an ARM cpu?”

    Uh, the power button or shutdown -h now? You need a link for that?

  130. wizard emiritus says:

    “System administrators keep the utilization at a reasonable level and turn off the chips not needed at certain times of the day.”
    Interesting. The only time I saw System administrators interacting directly with cores this was with an IBM Power Mainframe, Where they had the ability to add 10th’s of a CPU as needed to an LPAR.

    How about a link showing this done in the case of an ARM cpu?

  131. DrLoser wrote, ” On a 24-core system you are (assuming the same load pattern) using 2, maybe 3 or 4 cores (I’m being very generous here) and the other 20 are basically a waste of space.”

    Utter nonsense. These systems have huge numbers of processes and there is a wait-queue for each core. System administrators keep the utilization at a reasonable level and turn off the chips not needed at certain times of the day. On a modern system 24 cores can certainly be well-utilized.

    I could make the same argument for tires on a car. Where’s the need for four of them except to haul around the bloated weight of these gargantuan systems have four empty seats, an empty trunk and a 150HP engine running at 800 rpm in overdrive? Certainly four tires on a Solo would be utterly silly and would increase the air-drag, cost of maintenance, and price for zero benefit.

    Some arguments make sense. Others do not. We could all go back to 16-bit single-core CPUs but we don’t for good reasons. Neither must we stop at four cores even for small systems. If one does the maths, more and smaller cores does reduce the power-consumption for the whole system. This is a perfectly good compromise against system throughput/speed in many cases like my desktop and server.

  132. DrLoser says:

    Then again, Fifi, you mention the terrific advantage that the ARM Cortex-A53 has over, say, the Intel Haskell architecture: L1 cache of 32KB (not “similar,” but actually “identical,” you innumerate fool), but oh so many cores just waiting for a new thread.

    Let’s examine this little theory of yours, shall we? Call it the L1 theory.

    If “clearing the L1 cache on a context switch” were the sort of expensive catastrophe that you claim, Fifi, wouldn’t you think that Intel would compete directly against ARM by using the following simple piece of math?

    Intel has four cores. ARM has 32 cores. (I’m planning forwards for this great leap in motherboards.)

    Big problem! Mommy make the context switching not give ouchie!

    Ah well, there we have it. Thirty two divided by four is eight. Eight multiplied by thirty two is two hundred and fifty six.

    Don’t you think, Fifi, no wait, I know you are incapable of rational thought, but seriously, if you were Intel in this position … would it not occur to you to increase the L1 cache on the Haswell etc chips to 256K? Problem solved, and all that lovely silicon real estate to add things like GPUs and so on!

    Well,, no, obviously it would not occur to you. Nor do you understand the trade-offs in chip manufacturing.

    This is, I am afraid, because you are an imbecile.

  133. DrLoser says:

    Oh, and when I say “thrashing,” please try to think of it in terms of “thin clients” and computing.

    Not what your somewhat more obese clients offer up under the lamp-post on your more lucrative Saturday evening job.

  134. DrLoser says:

    Deaf Spy higher context switch rate also means higher risk of having to reload L1/L2 caches due to thread changes.

    I believe I can live with that on a 4 core machine with decent pipelining and a large swathe of the available real estate devoted to L2 cache, Fifi. Just as long as each core runs at 1.8 GHz rather than 1 GHz.

    On a general-purpose system, what you lose in context switching is more than made up for by:

    (a) Usage of all the cores. I haven’t looked it up recently, but on a 4-core system you are typically using 2 cores, with the other two as spares. On a 24-core system you are (assuming the same load pattern) using 2, maybe 3 or 4 cores (I’m being very generous here) and the other 20 are basically a waste of space.

    (b) The fact that 1.8 == 1.0 * 1.8.

    This is not a difficult equation to comprehend, Fifi; otherwise I would not advert your attention to it.

    You have to lose an awful lot of CPU/cache time in context switching before you come anywhere close to that ratio. At which point you are probably thrashing on any system whatsoever.

  135. DrLoser says:

    24 cores arm does make a quite good arm image build server with integrated qa. KVM does work on it. So it can roughly replicate a raspberry pi and the like.

    Ah yes, the world is desperately in need of a $1,250 motherboard that can “roughly” emulate a Ras-Pi.

    So you can think of it as 6 raspberry pis in a box.

    Funnily enough, Fifi, that is precisely what all of us here think of it as.

    Not, I should say, in a particularly complimentary way. Nor in a particularly cost-efficient way, should you be sufficiently out of your skull to want six Ras-Pis in a box.

  136. oiaohm says:

    Deaf Spy higher context switch rate also means higher risk of having to reload L1/L2 caches due to thread changes.

    So in particular work loads like LAVA you do gain from having more cores even if overall I/O and RAM speed is little lower

    Something you have not considered is everyone one of the 24 cores has the same size l1 cache as intel processes. So more cores to spread tasks over with tasks pin to cores lot lower need to completely flush the caches.

    http://armdevices.net/2017/06/27/socionext-24-core-arm-server-synquacer-sc2a11-60-lower-power-than-intel-at-same-performance/

    Deaf Spy if you are not needing more performance than the SC2A11 can provide its highly power effective for performance. Its also not that horrible slow thinking people had LibreOffice running on raspberry pi 2 running at 900 mhz. Biggest issue Raspberry PI 2 was not running out of cpu but was running out of ram and Raspberry Pi 2 was only 4 cores of A53 at 900mhz. Also the ram supported on a SC2A11 is faster than either a raspberry pi 2 or 3.

    Of course I would not expect SC2A11 to-do any good gaming. But general web browser and word processing its not going to perform badly and will improve as more of these support multi threading. Particularly if you are comparing to the performance an old system will provide.

    The price of SC2A11 systems I do question. SC2A11 are great for performance vs power with suitable workload but does the power saving in a home workload cases justify the extra payment most likely no. Part of the problem here is chicken and egg. You need enough volume of production to bring individual unit price down.

    There should be more of these multi core systems appear.

    https://fuse.wikichip.org/news/686/esperanto-exits-stealth-mode-aims-at-ai-with-a-4096-core-7nm-risc-v-monster/
    This 16 full out or order risc-v with 4,096 more optimized cores. There are more arm64, mips and risc-v chips due out this year in the 16+ primary core class. So I do wonder when Intel will start increasing number of cores in desktop machines again.

  137. Deaf Spy says:

    I see Beast running 200+ processes all day long: databases, web-servers, DHCP servers, I/O, commands, applications…

    You surely now that all these are most of the time in waiting state, consuming no CPU resources whatsoever, are you? Of course you are, as you bring a peculiar metric on the line.

    Ah, context switches… Obviously, Robert, you analyzed all metrics – CPU time, I/O time, RAM loads, just to mention the three most important. You extracted the data for all these, including number of CS over a period of time, including the CPU time spent on CS, calculated the mean values. You did all this, right? And now you are sure the CS are what slows your ancient system down. Having 4K context switches for a single operation surely sounds like hell to you. Not the slow IO, not the slow RAM, just the damn CSs.

    How a few CS less will compensate for the slow I/O and RAM is beyond me, but hey, feel free to contribute. You may start by explaining how asynchronous IO operations work.

  138. Grece says:

    A fool and their money, is soon separated.

  139. Deaf Spy wrote, for the umpteemth time, “why do you blindly believe that 24 cores is better than 1?”

    I’m not blind. I see Beast running 200+ processes all day long: databases, web-servers, DHCP servers, I/O, commands, applications… Most of these processes do not require blinding speed except in the context-switch. More cores certainly does help that. At the moment, Beast is just idling and there are 1K context-switches happening per second. Here, I’ll go look for recipes with “cheese” in the title. Oh, my, context-switches rose to 4K per second. That’s just pleasing me. Imagine my living room full of visitors discussing pizza or birds or veggies or …

    Wanting more cores is not about multi-threaded programming but processes here. I have a lot of processes running. 24 cores would also help in building software on GNU/Linux but I don’t do much of that these days. TFA may be about a developer’s box but it can also be a fine desktop box too.

  140. oiaohm says:

    A 1-GHz A53 is so slow in single-threaded performance that it hurts. Even single task will run slowly. With 24 cores, theoretically you may get some advantage in very specific tasks like complex parallelizable calculations, video processing, but basically nothing else.

    It is called a development machine because it is useful for something else Deaf Spy.

    24 cores arm does make a quite good arm image build server with integrated qa. KVM does work on it. So it can roughly replicate a raspberry pi and the like.

    So you can think of it as 6 raspberry pis in a box. Yes supporting 16 megs of memory means you can in fact do like 6 KVM with exactly memory match 1 and 2G memory boards.

    Something else to consider is a 53 is a in-order so it does not suffer from all the Spectre and the like problem. So it single threaded performance is not as far behind as it was due to the overhead on arm working around Spectre.

    Deaf Spy a 1G network file server at worst does not need more than 8 cores.

    As a software development environment for building images for arm and running Automated quality control for arm the mix of that Developerbox makes sense.

    Its a powerful enough system to run LAVA with real hardware boards and internal KVM instances. Yes the single threaded performance of a A53 might be bad but when you attempt qemu arm on a x86 its also very bad.

    So that box makes absolute sense if you are developing for arm boards and want a QA server box. For more general tasks has some issues. LAVA can be massively Parnell so can hold those 24 cores at full load.

  141. Grece says:

    For the price of this one POS machine, I can easily build two decent NAS machines, that would be infinitely scalable.

  142. DrLoser says:

    I’d recommend either Bash programming or Pascal programming, Robert. You are a self-acknowledged expert in both.

    Can’t quite bring to mind how you’d multi-thread anything in either one of those, however.

  143. DrLoser says:

    And, obviously, 24 cores are better than 1 because you can turn them up to 11.

    I have seen the light!

  144. DrLoser says:

    In point 3 I envisioned a case where the server has a high-performing IO bus and has to do some quick and simple transformations of data. Something like Azure IoT Edge. Definitely not full workflow, and definitely no storage. Some data comes in, gets processed lightly, and pushed forward to the next node.

    I think we’re both on the wrong board if either of us believes that a meaningful discussion is in any way likely.

    But before the cascade of gibberish falls upon us: yup, why not? I wouldn’t pay quite as much as $1,250 for such a thing, because quite honestly it strikes me as a 2018 version of a NAS, but then again … the results are going to be pushed forward to the next node. For another $1,250, unless you do the sensible thing and go for AWS or Azure or whatever.

    I think we can both agree that Robert is simply whining about the cost of his perfect, yet perfectly unsuitable 24 core server … when all he really needs to upgrade the Beast is a $300 NUC plus 4GB RAM plus 250GB SSD.

    How about that as an interim option, Robert, whilst you grow ever older and need to save up for your “dance card?”

  145. Deaf Spy says:

    Dear Doctor,

    For the sake of a meaningful discussion, I’d like to dig a bit more into the topic.

    In point 3 I envisioned a case where the server has a high-performing IO bus and has to do some quick and simple transformations of data. Something like Azure IoT Edge. Definitely not full workflow, and definitely no storage. Some data comes in, gets processed lightly, and pushed forward to the next node.

    But, Robert, don’t let this distract you:
    Do you have any experience in multi-threaded programming, Robert?
    If not, why do you blindly believe that 24 cores is better than 1?

  146. DrLoser says:

    Which thought actually brings up a very interesting question, in therms of Robert’s competence as a programmer.

    Multi-threaded programming (say, POSIX) has been mainstream since, I think, 1990 or so. Multi-tasking programming was around before then (say, with Tandem or Stratus): say, 1985 or so.

    Do you have any experience in multi-threaded programming, Robert?

    If not, why do you blindly believe that 24 cores is better than 1?

    You may quote from “Animal Farm” if you wish.

  147. DrLoser says:

    This means that it is good as:
    1) a file server, or
    2) a first interface for IoTs, which sole purpose would be to collect incoming requests and queue them into an event bus, or
    3) Processor for queued events for very, very limited processing (like, basic transformation of the received package and shove it down further the pipe to the next layer)

    I would respectfully disagree, Deaf Spy.

    1) Sans special extra hardware, it’s going to be rubbish at being a file-server. Well, you could do a little better by dedicating it as such and tuning up a full Windows Server 2016 OS with full-bore asynchronous I/O and fibres … but, nice as that might be, there are several implausibilities. The very least of which is that you wouldn’t use it for anything else at all. So, another $1,200 for the next device.

    2) Unless your local IoT environment involves about 10,000 “things,” there seems little point in such an extravagance. But, hey, if you want to queue up lots and lots of (TC/UD)P/IP requests, once again, 24 cores ain’t gonna help much, unless you have 24 dedicated ethernet sockets. Which seems unlikely. You could always use work-stealing, I suppose. Or, again, Windows Server 2016.

    3) You don’t “queue” events over 24 processors/cores. You know this, I know this, and Robert “Mr Moore’s Law” Pogson doesn’t have a clue. You use one (preferably) or two of those cores to handle the I/O, and you rewrite practically everything you do in terms of parallel processing and very careful memory management and pipelining and so on — I suppose a NUMA architecture might help — and add a bit of Rust here, a bit of lazy evaluation there, a very highly tuned database, and possibly even a dash of lock-free inter-process communication.

    What strikes me is that Robert’s basic position here is:
    a) I am helpless, O Lord.
    b) Save me, for I am more worthy than everybody else!
    c) Give me power beyond my wildest dreams!
    d) What, $1,300? You’re joking, O Lord. Maybe if I’d bought Microsoft stock instead of hoarding gold, but that’s a bit rich for my blood!

    All very sad, really.

  148. Deaf Spy says:

    You don’t even understand what this board is designed for, Robert, do you?

    A 1-GHz A53 is so slow in single-threaded performance that it hurts. Even single task will run slowly. With 24 cores, theoretically you may get some advantage in very specific tasks like complex parallelizable calculations, video processing, but basically nothing else.

    See your own source:
    “it supports from cloud computing to edge computing, which is the key to IoT era”
    This means that it is good as:
    1) a file server, or
    2) a first interface for IoTs, which sole purpose would be to collect incoming requests and queue them into an event bus, or
    3) Processor for queued events for very, very limited processing (like, basic transformation of the received package and shove it down further the pipe to the next layer)

    If you build a home server for thin clients on top of this one, you will be terribly disappointed. But don’t worry, you never will.

Leave a Reply