Decline And Fall Of Wintel

Confirming what we already knew, “Worldwide PC shipments totaled 82.6 million units in the fourth quarter of 2013, a 6.9 percent decline from the fourth quarter of 2012, according to preliminary results by Gartner, Inc. This is the seventh consecutive quarter of shipment decline.”

It’s about time that retailers realized that consumers want small cheap computers. They love Android/Linux on mobile PCs of all kinds. They would love GNU/Linux too if offered a choice, even on legacy PCs. If OEMs and retailers want to beef up their numbers, they should start shipping more GNU/Linux systems in 2014. 7 consecutive quarters of decline in Wintel PCs should be sufficient evidence even for the most stubborn fans of Wintel. The problem is not with the kind of Wintel systems retailers are offering. The problem is that they bear the burden of Wintel. If retailers can make a little money selling Wintel PCs at ~$300 per box, they could make a lot more money selling GNU/Linux boxes for half that. The boxes don’t need to be so large. The PSUs don’t need to be so large. The CPUs don’t need to be so powerful. Get on with the business you’re in, selling IT, not selling M$’s crap.

See Gartner Says Worldwide PC Shipments Declined 6.9 Percent in Fourth Quarter of 2013.

See also, What a surprise! 2013 was a lousy year for PC sales

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.

91 Responses to Decline And Fall Of Wintel

  1. ram says:

    Chrome works great for watching YouTube educational videos, but I don’t trust it for anything else.

    Separately I note, Intel is divesting of its “cloud media” technology. Perhaps the future of “cloud” is not so great after all.

  2. dougman wrote, “adding in Chromebooks and Chromebases”.

    Off-hand, I would say Google Chrome browser is a little too “chatty” with Google for use by a high-security outfit. I can see some variant of FireFox working for them, though. Chrome is designed to give top performance and to feed information to Google, neither of which is the top priority of security-minded folk. In principle, Google can commit a certain farm of servers to them and wall it off from the world, but Chrome just babbles on to Google about what links are on the page and pre-fetching them. That’s quite insecure. One layer of security might involve fetching certain documents or logging in in a narrow window of time. Pre-fetching would mess that up.

  3. dougman says:

    M$ loses to Google in Maryland.

    “Previously, each agency ran its own email servers — from Microsoft Exchange and Novell, to in-house platforms. We knew to move these disparate email systems into the cloud would decrease complexity and improve intra-agency collaboration, but any cloud-based solution we selected had to meet high security standards.”

    http://googleenterprise.blogspot.ca/2014/01/the-state-of-maryland-goes-google.html

    http://googleenterprise.blogspot.com/2013/02/us-naval-academy-goes-mobile-with.html

    So much for needing Windows, now they should dump the Win-Dohs OS and start adding in Chromebooks and Chromebases.

  4. dougman says:

    “As tablets continue to increase in popularity and eat into laptop sales, it’s also possible that the industry has hit a spot where tablets are now able to replace laptops for more and more people. “We’re arriving at a sweet spot where tablets are now becoming desktop, let alone laptop, replacements,” says Miller. “These tablets are combining the right mix of portability / weight, performance, and price such that a lot of users who would have held out for a laptop, because it was more powerful, might want a tablet instead, and even be willing to live with some compromises in order to benefit from the other form-factor benefits and flexibility.”

    http://www.theverge.com/2014/1/9/5291910/microsoft-ces-2014-nowhere-to-be-found-report

  5. oiaohm says:

    Deadlock you say dies bigger. This is where you have a big problem.

    Take a current day Armv7 Codex A15 Quad core. Now when you make an Codex A57 at current day nm. A quad core Codex A57 fits into every A15 with space left over.

    So a 16 core Codex A57 is not large. A57 is 40 Percent smaller at the same nano meter size. A15 are nm 32. A57 are nm 14 to 20 in production. Yes A15 if you could take it down to nm14 would fit 4 cores in the same space as 1. So there is +40% left over for cache and other things.

    There is no requirement for companies making arm chips to reduce the number of chips per wafer any time soon. If anything increase number per wafer.

    Deadlock the intel chip you talked about giving armv7 trouble is 14 nm and its die size is larger than a quad core armv7 cordex a15. We are really talking with armv8 where 1 armv7 core gets replaced by 4 armv8 with cache and other things. 1 armv8 core being 3 times faster at the same clock speed is a problem for intel. The fact that 4 go in the same area.

    Yes the insanity of a 16 core mobile phone could come.

    Deadlock so what is a quad core arm today becomes a 16 core arm by the end of 2014.

    Arm moves in quite huge leaps. So 3 times faster due to Cpu redesign. 4 times by nm alteration. So 12 times as much processing power without touching the clock. So this is why I am laughing when you say an armv7 is only 1/3 the speed of x86 chip. That is no where near enough to stay ahead of arm next generation. Yes next generation chips you can get on prototype boards now. Even the power usage of the x86 is nothing to write home about.

    Note the reduction in nm also reduces power requirement. The armv7 and armv8 public were both produced at 32 nm. Yes the 14nm armv8 draw less power. Fairly much that 4 armv8 cores draw the same as a armv7. So a armv7 quad core has the power usage of a armv8 16 core.

    Deadlock you also presumed the number of cores in arm would result in scale increasing power usage. The problem is this is not the case either.

    Intel just got to 14 nm first. Intel has to attempt to get to 7nm to attempt to stay ahead of arm or completely redesign x86.

    There is going to come a point where there is no lower nm for intel to run to.

    This is the problem next generation arm. Less die per core. Wayless. Less power draw per core due to redesign. Less power draw due to lower nm. Higher performance due to higher clockspeeds and redesign.

    So armv8 vs x86 will be a battle.

    Deadlock if arm was making their dies bigger they would be taking about a CPU with more than 16 cores. There is chat about 32 core chip this being 16 cores A57 and 16 cores A53. A53 is serous-ally small.

    Deadlock really get how you were tricked yet. Intel put latest generation of their chip against arm a generation behind.

  6. Deadlock wrote, “Since when is a *third* of the performance “close enough”?”

    Since reaction time is the blink of an eye. No one but Deadlock cares if something happens in 1/3 of a blink of an eye.

  7. oiaohm says:

    Deadlock the problem here toms hardware and other reviewers you have access to only do public provided hardware. Armv8 chips are in production currently only for selected clientele. Yes two of those selected clientele is amazon and facebook. You want benches on them you have to know their staff.

  8. oiaohm says:

    Deadlock arm caches on armv8 are in fact different design to x86 in a big way as well.

    In an armv8 L3 is fully shared between at least 16 cores. L2 is fully shared between 4 cores. Again this is arm doing different to X86.

    I’ll say it again – for ARM to get the same performance as x86 = more heat, more power, more expensive. The ARMv8 core may get smaller, but the number of cores and amount of cache needed if they’re serious about taking on Intel in the desktop and server space will make dies bigger.

    Again I say you are a Idiot on this topic. Arm has looked at everything x86 chip makers have done. They have the advantage of seeing everywhere Intel and AMD stuffed up their designs on real workloads. L2 shared between 2 cores only that is a mistake in real world workloads.

    Please note a armv8 is 3 times faster than an armv7 at the same clockspeed. You have benchmarks of armv7 saying that they are 1/3 of a x86

    Deadlock sorry you brought the very benchmark that brings you undone. Then you have to remember Cortex 57 and 53 run at faster clockspeeds. There is currently an arguement in server space if 57 or 53 will be used. 53 is missing a few cpu features but is smaller again to a 57 and uses less power.

    Deadlock we are reaching the point where we cannot keep on going to lower and lower nm in silicon production. Each step smaller is starting to threaten to cost in the trillions to make the factories.

    Size vs Performance of the cpu core is going to become critical. Same with size vs performance of the gpu.

    Deadlock you are not Facebook or any other big data centre. They have armv8 chips. They are all saying the same thing they are looking to go arm. Yes armv8 chips do compette well against top intel chips.

    Here is something else where deadlock is a idiot. Why does an armv8 not need as much cache as a x86 chip. L2 and L3 can be compressed data in an arm chip. Yes L2 and L3 are different design in arm to x86. x86 does not include a flag to say this page in cache is compressed arm does. Every point in cpu design arm does more with less.

    Deadlock armv8 has the issue of more than double the cores in the same area of silicon at the same nm of production as a x86.

    Deadlock there is not a single section of a arm chip design that maps to a x86 chip design. You cannot say that arm needs X as big as the x86 chip. Reason arm design is just that far different. Arm chips are design todo every trick to reduce silicon required.

  9. Deadlock says:

    Deadlock blathered on about irrelevant stuff… My Phenom idling is easily beaten by many ARMed processors working hard.

    You accuse me of blathering about the irrelevant? Pot, meet kettle!

    Your argument is like saying you’re a faster runner than Usain Bolt, but only as long as Bolt is standing still. That argument summed up in one word – “stupid”.

    The benchmarks you cite are different benchmarks run by different organisations. You can’t mix and match – what was tested would be different. Did you also note that the Core i7 you cite on the Tom’s Hardware site is the FASTEST processor in that particular test? Where’s the $50 ARM that’s “almost as good”?

    On the OpenBenchmarking test, the i7 does the LAME test in a third of the time of the ARM. How does this prove ARMs are “almost as good” as an i7?

    As usual, you’ve posted sources that disagree with you.

    That’s close enough that few consumers would notice in the operation of their systems.

    Since when is a *third* of the performance “close enough”?

    The ARM with equivalent throughput to a Phenom II is?

  10. dougman says:

    Dreadcock, sure has a hard on in trying to mark Pogson as being technically inept; arguing over a i7 vs a Arm processor is just silly. I am sure Dreadcock knows precisely what he is talking about, secretly I think they are trolling for Intel *Rolls-eyes*

    Concerning such, namely the i7, is built on older architecture built for performance disregarding power, the other the Arm, was built for energy efficiency like by a factor of 15.

    In this day in age, Arm processors are perfect for data-centers servers and those looking to lower their PUE.

    “The onus is now on IT hardware and software. The hardware manufacturers have gotten the message already and have been addressing it in several ways. The first and most directly and by its very nature starts with the CPU. While Moore’s Law continues to prove out with ever rising performance it comes with a price – more energy and heat. The CPU manufacturers are well aware that they have been improving the performance-per-watt ratio for the last several generations of chips. In fact, it is the mobile device market (laptop, tablet and especially the smartphone), that has done more for chip energy efficiency (to deliver better battery run-time), rather than the data center market (which up until recently did not care that much about energy usage). However, they are near the physical limits of existing technology – unless and until they can make the quantum leap via the “next gen” chip material (i.e. moving past silicon based CPUs). 2014 may also be the year where low power processors (ARM, Atom, etc.) start making significant inroads in the hyper-scale data centers.”

    http://www.datacenterdynamics.com/focus/archive/2014/01/more-outsourcing-less-carbon-better-efficiency-2014

    http://www.forbes.com/sites/timworstall/2013/12/15/now-facebook-is-looking-at-dumping-intel-for-arm/

    MS is not happy of course. While Windows PC market is declining, the traditional MS partners are embracing Linux powered Chrome OS and Android. We have seen OEMs like HP and Lenovo offering all-in-one Android PCs and now Intel has also joined the bandwagon. That’s going to hurt as there was a time when the two companies monopolized the market and gained nick name Wintel – though it was an abusive monopoly and not at all healthy for market.

    So to summarize, Arm is making serious in-roads and Intel is playing catch-up. The Wintel monopoly has tumbled.

    http://www.businessinsider.com/wintel-monopoly-2012-9

  11. Deadlock blathered on about irrelevant stuff… My Phenom idling is easily beaten by many ARMed processors working hard.

    Here is a LAME encoding benchmark. My Beast would be 180s, Intel Core i7-4770K (Haswell 4c/8t) 3.5 GHz (Turbo 3.9 GHz), DDR3-1600, 1 MB L2, 8 MB L3, would be 79s.

    A benchmark of Core i7 at 3gHz on 4 cores gives 17s for LAME encoding.

    Here’s one of ARM quad-core at 1.8gHz doing 51s.

    That’s close enough that few consumers would notice in the operation of their systems.

    My Phenom runs a 90W. Any ARMed CPU would beat that.

  12. Deadlock says:

    Nonsense. Look at the specs for any of this year’s multicore 64bit ARMed CPUs. They have clock-speeds over 2 gHz.

    Name one that beats your Phenom in benchmarks, then.

    Name one that gets remotely *near* a six-core i7 in benchmarks.

  13. Deadlock wrote, ““Beast has 4 cores at 2.5gHz. There are ARMed CPUs today with that much throughput.”
    You made claims with no basis in reality”

    Nonsense. Look at the specs for any of this year’s multicore 64bit ARMed CPUs. They have clock-speeds over 2 gHz.

    e.g. seeMediaTek announces first true octa-core ARM processor: “Perhaps the most critical piece of information about the MediaTek MT6592 is that it uses all Cortex-A7 cores for CPU tasks, as opposed to Cortex-A15 cores seen on its competitors. While the A7 cores have the same architecture and feature set as the A15 cores, the A15s are around twice as powerful: A7 cores deliver 1.9 DMIPS per MHz, while A15 cores can deliver up to 4.0 DMIPS per MHZ. This means that in applications that only use a few of the processor’s cores, an A15-based chip has a significant advantage in processing power”

    So, the ARMies are producing multiple choices competitive with Beast. Do the maths. 4 DMIPS per MHz X 4 cores X 2.3 gHz or so is a Hell of a lot more throughput than I use from Beast except when building kernels. I have the time for that down to less than 10 minutes by not building stuff Beast doesn’t need so I don’t care if it takes a bit longer on some other device and I can always build cross-platform.

  14. Deadlock says:

    That’s irrelevant! My Beast is idling 99.9% of the time.

    Who cares?

    YOU described an ARM CPU as “almost as good” as a $1000 top-end Core i7.

    YOU wrote – “Beast has 4 cores at 2.5gHz. There are ARMed CPUs today with that much throughput.”

    You made claims with no basis in reality, and after being asked for proof, you claim it all irrelevant, wave your hands, shift the goalposts, try to change the subject, and attempt to smear anyone who only wanted you to back up your claim with facts.

  15. Deadlock wrote, ” There isn’t even an ARM that beats your Phenom II, and you claim an ARM could take on a six-core i7? “

    That’s irrelevant! My Beast is idling 99.9% of the time. So are a lot of desktop systems. Most systems are sluggish not because of the CPU but because of malware or software bloat or I/O. The CPU has nothing to do with that. Users are quite happy if stuff happens in the blink of an eye. At thousands of MIPS that happens whether it’s an ARM or Intel CPU. Price/performance matters and most users of PCs get sufficient performance at a reasonable price from ARM. I can buy a whole box for the price Deadlock mentions for one CPU from Intel…

  16. Deadlock says:

    Cheaper, for sure, the others one way or another. You can’t stick a Xeon chip in a smartphone.

    And you wouldn’t stick an ARM in a server, desktop or workstation that needs a lot of CPU throughput.

    You’re the one that said there are ARMs are equivalent in *throughput* to quad-core server-class x86 chips. A claim you still have not proved.

    Meanwhile, that Intel chip uses 130W, 22W per core, and the ARMed chip uses ~13W per core…

    You do know there are $300 quad-core Core i7s that run at a maximum of 35W (~9W per-core) and would stomp any ARM in existence? Just checking.

    Let’s assume that ARM heads down the path oiaohm says it will and will have many more cores (double and more) than the Intel. For the sake of argument, we’ll start at double, and the i7 you cite has six. Twelve cores at 13W = 156W. And it *still* won’t be faster than the Intel.

    So you can pay ~$1K for “state of the art” or you can pay ~$50 for “almost as good”. Your choice.

    But it’s not “almost as good” – it’s nowhere *near* it! There isn’t even an ARM that beats your Phenom II, and you claim an ARM could take on a six-core i7?

    Please provide a performance benchmark result where an ARMv8 is in the same SPORT as any Core i7, let alone the same league.

    Where do you get your $50 figure, by the way? I cannot find any pricing for ARMv8 chips. Prices for devices they’re going to go in, but not the chips.

    I’ll say it again – for ARM to get the same performance as x86 = more heat, more power, more expensive. The ARMv8 core may get smaller, but the number of cores and amount of cache needed if they’re serious about taking on Intel in the desktop and server space will make dies bigger. And don’t assume Intel are sitting on their thumbs – performance in the new Atoms outstrips ARM (where benchmarks are available), and power and cooling needs of Atom will only get lower.

  17. oiaohm says:

    Also note armv8 is 3x faster than armv7. So what took 8-9 armv7 takes 3 armv8. Yes 3 armv8 cores almost fit in the space of 2 armv7 cores at same scale production.

    Silicon requirement of arm cores is less. x86 is required to use more expensive and complex fabs to use the same amount of silicon area as a arm chip.

    Yes 20nm is way cheaper to produce than 14nm. Future is transistor efficiency. Arm designs had fairly good transistor efficiency. Armv8 improves that massively.

  18. Deadlock wrote, “What are ARM’s advantages over x86 again? Low power, less cooling needed, and cheaper?”

    Cheaper, for sure, the others one way or another. You can’t stick a Xeon chip in a smartphone. Atoms still have x86 baggage and no matter how Intel optimizes Atom, Atom will cost more to produce than ARM. At the moment, Intel needs 2X the resolution just to be competitive with ARM but the price is the killer. A state of the art ARM chip costs less than $50, has more CPU throughput than an Xbox, wonderful graphics, and uses only 5W while an Intel chip costs ~$32. Further, the Tegra K1 is competitive in graphics performance with some of Intel’s mainstream notebook chips so there’s lots of leverage that Intel is feeling. That’s why Atoms exist.

  19. oiaohm wrote, ” So they need less transistors to perform the same operations.”

    One can do benchmarks to test that:
    Toms Hardware claims Core i7 4960X @ 4gHz gets ~190 GIPS, about 32 GIPS per core and 12 IPS per core per clock cycle.
    Wikipedia shows ARM v8 gets 4.76 IPS per core per clock cycle.
    Meanwhile, that Intel chip uses 130W, 22W per core, and the ARMed chip uses ~13W per core… So you can pay ~$1K for “state of the art” or you can pay ~$50 for “almost as good”. Your choice. Oh, and Intel needs 14nm to get its performance. ARM is still at 20nm or so…

  20. oiaohm says:

    Deadlock I should have started why x86 servers stops at 8 cores. You find out at 12 and 16 cores x86 you run into the annoying case your performance is not increasing.

    Yes you spend a high price on but you are not getting benefit.

    AMD Shared double-sized Flex FP is a bug bear if you are truly doing a lot of floating point. Yes that 16 core AMD x86 can be tied up so much in internal locking attempting to get floating point its only a 8 core because only have the cores are getting to progress.

    Intel 12 and issues with broken up L3. So cannot load balance tasks effectively.

    Deadlock both AMD and Intel have major issues in their x86 designs. Arm really does have the advantage coming last. They get to avoid what the others have done to themselves.

  21. oiaohm says:

    http://www.anandtech.com/show/6420/arms-cortex-a57-and-cortex-a53-the-first-64bit-armv8-cpu-cores

    Deadlock the bench data is known on armv8.

    cordex a15 and cordex A7 is armv7
    codex a57 and a53 is armv8

    The simple answer you are bogus Deadlock you don’t know Arm cores. You are making a stack of presumes that don’t match up to the way Arm is designing their chips.

    Armv8 don’t increase power they reduce it. So out of order don’t cost arm any more power. Larger pipe lines also don’t cost arm any more power. Why arm found another way to make space for those features.

    Notice the artical it was published 2012. DeadLock is 2 years out of date on the trash he is quoting.

  22. oiaohm says:

    Deadlock Armv8 cores is not bigger than Armv7 cores. Armv8 cores is a lower transistor count than Armv7 cores .

    Your logic about will make arm cores bigger is incorrect.

    Deadlock arm has done something to make the transistors do more roles. This is very custom design software.

    Deadlock arm and intel design of cpu is vastly different. Armv8 current design is close to fpga. The internals of the chip are altering configuration on fly.

    As its put is a method of reducing dark silicon. That is silicon that is not powered on.

    Armv8 includes some very interesting tech. Tech that is not in x86 processors.

    Armv8 also generate less heat than Armv7 and draws less power than a Armv7. Its all due to the one bit to tech.

    Deadlock there is two ways to face the end of Moores law. 1 like arm use transistors more effectively for more roles or 2 like intel pray it don’t hit too hard.

    Arm patented the reusing system they have where they overlap their circuit designs. So they need less transistors to perform the same operations.

    Deadlock reduced transistors make armv8 simpler to make than armv7.

  23. dougman says:

    Whats the matter, http://tmrepository.com/ not giving you anything worth discussing?? Seems the discussions have gone downhill over there, perhaps the site will die off. You blokes have reverted to arguing with each other…LOL.

    Meanwhile, I am glad that MicroSh1t does not have this power over me: http://www.dailydot.com/technology/tor-botnet-microsoft-malware-remove/

    Of course I am not condoning botnets or other nefariousness, but the general idea of an entity reaching into your ‘home’ and removing something is downright NSA like.

  24. Deadlock says:

    @oiaohm

    Amd x86 server cores stop at 8 cores.

    Not true.

    AMD have 12- and 16-core CPUs already. (Opteron 634x, 637x and 638x). Intel also do 12 core XEONs (E5-2697v2).

    Notice something if there is a 30 percent performance difference will it really matter when you get double the number of cores with arm?

    I have already pointed out that with an ancient cache-limited dual core AMD, it took EIGHT recent ARM cores to beat it.

    Yes its not going to be quad core vs quad core. Less silicon space of arm cpus cores equals more cpu cores per chip.

    Features such as out-of-order execution and larger pipelines that are being introduced on ARM will make ARM cores bigger. More power needed, more heat generated. More ARM cores per-package means more power used, more heat generated, on an exponential scale.

    All these increases in CPU complexity for ARM CPUs will make them more expensive to produce – see Moore’s SECOND Law. Less chips per-wafer is only the beginning of the hurdles here. ARM themselves dont have to worry about that – they only license out their designs and patents – but the companies fabricating the chips will have to ramp up their costs.

    What are ARM’s advantages over x86 again? Low power, less cooling needed, and cheaper? Those advantages will start to disappear if ARM go after the desktop, while BayTrail will start making inroads on ARM’s advantages while still retaining x86 compatibility.

    ARM are going to run into all the hurdles Intel did with x86 – the ARM architecture is not CPU pixie dust. They’re subject to the laws of physics just as much as Intel or AMD.

  25. Deadlock says:

    “Perhaps I should block all comments so I don’t have to read crap like that. ”

    Perhaps you shouldn’t make claims that have no basis in reality, so people have no cause to question you. Then none of us have to read crap.

  26. oiaohm says:

    ssorbom the correct answer it is been done now arm but I personally without the right workloads I would not. It will become clear as I explain the limitations of armv7 vs armv8. I am basically talking from real world examples.

    Issues are a bugs in armv7 has become very clear. You do need to use armv7 big little to get power effectiveness that is really a pest to have the Linux kernel get the switching correct. So it balls it up. A balls up arm never really notices on phones. Idle armv7 is perfect at load it has issues. Now x86 not perfect at idle does power management better at load. Yep heavily loaded x86 can win lightly loaded arm wins. That changes with armv8 that unbelievably better under-load. Between armv8 1/3 to 1/4 less power usage than armv7 in the CPU of an armv8 at full load even when it running faster than an armv7.

    All the Armv7 are all still using odd ball GPUs that are not common desktop usage GPUs with no pci-e to connect more normal video cards. Yes Armv7 chips totally don’t support pci-e so forget hooking up a standard video card. Armv8 one of the big feature changes is pci-e. Armv8 has Pci-e and Mobile Pci-e support.

    Now even with all those limitations the armv7 desktop and server devices still have there places. But limited places.

    Yes Armv8 lands second half of 2014 as mass produced chips. Then we will find out if Nvidia or Amd will support there high end cards on Armv8. If they do what we know as the desktop could change massively.

    We are inside a year of a full out war for desktop and server space between arm and intel.

    Yes accelerators for video work and the like also require pci-e slots.

    As I said the work loads you were talking about were CPU and GPU. Arm has been highly crimped on GPU access. Some of this is arm design issues lacking the systems to connect the GPUs so forcing usage of SOC GPUs that mostly don’t perform well.

  27. ram says:

    Beware of quantum ducks: quark, quark 🙂

  28. ram says:

    I have a number of machines (nodes in a cluster) that have far more GPU power than CPU “grunt”. Yes, the CPU’s are AMD64 architecture but their clock rate and number of cores hardly makes a difference to most audio and video rendering/transcoding tasks. In fact, those machines are all I/O bound. If an ARM processor was used instead of AMD64 it would not make any significant difference.

    Haven’t had a chance to benchmark an Intel Phi board yet. Those are interesting, they are a “cluster” of Atoms (AMD64 architecture) acting as a GPU and OpenCL board. And now Intel has the Quark chips (they are 32 bit now, but a 64 bit one is coming). Curiously a big user of Intel Phi boards is the theoretical physics supercomputing community. It’s all on Linux, of course!

  29. ssorbom wrote, of an ARMed PC (not smartphone nor tablet…)“where could I find one of these machines on the market?”

    They’ve been on the market for years as desktop units like TrimSlice. Dual-head 1080p, dual gigabit/s Ethernet, Wifi AND Bluetooth, 5.3in x 3.9in x 0.8in box and it runs Ubuntu GNU/Linux OR Android/Linux, up to 2 gB RAM and 32gB SSD storage, and 4 USB ports for keyboard/mouse. Price, $219. Beats the pants off many Wintel boxes. Of course, these days there are Chromebooks which run GNU/Linux on ARM. eg Samsung 11.6″. You can install another OS on it if you wish, like Debian GNU/Linux. You can buy one at Walmart for $243. That’s hard to beat, portable too… I think if Beast died tomorrow I would hook one up to my old keyboard and a monstrous NAS/DB/Web-server.

    etc. etc.

  30. ssorbom says:

    oiaohm,
    You seem to be saying that ARM can be made to do all the things I mentioned. Okay, the next question is, where could I find one of these machines on the market? I don’t mean a smartphone or tablet, I mean something with a decent keyboard, plenty of local storage, and a non-ROM OS. I have been looking since this discussion started, and everything I have found fails on at least one of those three counts. My ultimate goal would be something that can run Debian with standard KDE plasma desktop.

    I realize it is different to say “It can be done” as apposed to “it is being done now” but part of my argument is that ARM in the present day is still not a drop-in replacement for current PC technology. But you have proven me wrong before, so “knock my socks off” 🙂

  31. dougman wrote, “Talk about a pack of hypocrites!”

    Hmmm… Netcraft reports their OS is GNU/Linux but the server is IIS… I guess they welcome the lower cost of hosting GNU/Linux.

  32. dougman says:

    All the behind the scene action of the troll-pack may be viewed at http://tmrepository.com/stream/

    When you read their ‘About’ page, you begin to understand their thinking.

    Linux users = FREETARDS.

    When you visit the page, look to the lower right and you will find the words “Power by Django”

    Django, is a free and open source web application framework, written in Python. Talk about a pack of hypocrites!

  33. matchrocket says:

    “Now, remind me: Why again was it that I had to care about benchmarks?”
    Remind you, again? This is your first post. Unless you posted under a different nym and forgot.

  34. luvr says:

    May I humbly submit that I don’t care about benchmarks? Raw performance isn’t exactly the highest priority for me; instead, I want an affordable system with good performance.

    When I selected the hardware components for my current desktop computer, I opted for an “AMD Athlon(tm) 64 X2 Dual Core Processor 6000+” CPU. Far more powerful CPUs, particularly from Intel, will have been available at the time, but I considered them overkill. I didn’t even install a separate sound or graphics card, since the motherboard is more than good enough.

    I’m still using this system, and assuming that it doesn’t break down, I have not idea why I would ever want to upgrade it again.

    In fact, my dad is using an identical system (except for the hard disk space), and he all his non-linear video editing on it, without any real issues (except that he does still keep a Windows XP partition around, specifically for this use; he runs Windows XP with all networking disabled, though, and even for video stuff, he’s slowly but surely beginning to explore Linux). Granted, this computer isn’t the fastest possible option when it comes to rendering, but it is more than adequate.

    Now, remind me: Why again was it that I had to care about benchmarks?

  35. oiaohm says:

    CyaSSL is available for Windows, and yaSSL (the GPU port) is available for CUDA, therefore it’ll work in Windows.
    I should have been more exact. Even that CyaSSL is available for windows you don’t find firefox or anything else built against it out box.

    Android can have its core SSL switched to GPU by OEM/ODM. This is why benchmarking Androids is so hard. Different sections of Android API/ABI can be redirected into the GPU. This is becoming more common.

    Android a benchmark where you think you are testing the CPU can very quickly be testing the GPU instead.

  36. Deadlock says:

    http://en.wikipedia.org/wiki/CyaSSL exactly who would think your ssl can be running in opencl on the GPU. Of course windows does not do this.

    CyaSSL is available for Windows, and yaSSL (the GPU port) is available for CUDA, therefore it’ll work in Windows.

  37. Deaf Spy wrote, “Perhaps, you should try to find trustful sources you agree with, instead of causing confusion with sources you do not agree with.”

    Perhaps I should block all comments so I don’t have to read crap like that.

  38. oiaohm says:

    Deaf Spy a desktop without a GPU is a brick these days. That is the problem. Servers without accelerators or GPUs are also fairly close to bricks.

    http://hackaday.com/2012/12/14/leveraging-the-gpu-to-accelerate-the-linux-kernel/

    Yes serous-ally Linux kernel benchmarks can in fact be effected if you have a compatible Linux GPU and if the Linux kernel can use that GPU directly.

    Its every level in a Linux system finding ways to offload to GPU.

    This is the problem the major differences between same speed arm model cores running at same speed you see a lot is not the CPU. Its the GPU.

    Deaf Spy performance maths is way harder than you can dream.

    So its what CPU and what GPU and how well that integrates into the OS running on top of it results in the final performance.

    render-nodes coming to the Linux world will see more and more things using the GPU. Since render-nodes do away with having to open up a graphical window to use a lot of things.

    Deaf Spy in the browser example of performance not only the graphical elements are off loaded to GPU. The SSL decoding is also offloaded to the GPU.

    http://en.wikipedia.org/wiki/CyaSSL exactly who would think your ssl can be running in opencl on the GPU. Of course windows does not do this.

    Deaf Spy see the problem yet is very hard to get a clear benchmark of a arm cpu most of the benchmarks out there are cpu and gpu as one on arm with parts of the work load ending up in both.

  39. Deaf Spy wrote, ” I would only dare say that opening a web page is not dependent solely on the CPU. GPU has its major part, and this is why browsers aim for hardware graphic acceleration these days.”

    Users don’t care about CPUs. They care about the performance of the system. I know that opening a web page has many factors but the CPU is one of them and the systems tested show that ARMed systems do well. Some folks are spending many hours daily on the web and want good performance at a reasonable price. ARM clearly does that. I have seen ARMed systems that are a bit sluggish but most of them were years old. My smartphone, for instance, is a Samsung Galaxy S, 3+ years old. It has a single core and as slow as it is, I can use it to post to this blog or update my databases or find out where I am. That’s all I need it to do and its perfectly fine. It cost me nothing because the “little woman” cast it off for the newer BB Z10 which is much faster but crap as far as I am concerned because I have to figure out how to do anything with it. Time to load this page on that smartphone? Several seconds. Do I care whether an Intel could do it a second sooner? Nope. Folks would rather get a faster ISP than a faster CPU. Several seconds of my time is much more important to me than beefing up Intel’s bottom line.

  40. Deaf Spy says:

    Dear Mr. Pogson, this humble one only quotes your own source. I am unworthy to contradict Tom’s hardware on the topic. I would only dare say that opening a web page is not dependent solely on the CPU. GPU has its major part, and this is why browsers aim for hardware graphic acceleration these days. I would even have the recklessness to suggest your reliance on loading web pages as sole indicator for CPU power is ill-advised.

    Perhaps, you should try to find trustful sources you agree with, instead of causing confusion with sources you do not agree with.

  41. oiaohm wrote, “Most people don’t need more than a core2duo if they have a decent video card. A core2duo without a decent video card you are in trouble.”

    Good points above. The vast majority of us get along fine with whatever is in the system. In my home, the only system that struggles is an Atom with 1080p on a big screen. Beast, for instance has just a built-in video chip and many others are content with a core of the CPU doing the video. I agree with oiaohm, though, that anyone who needs vast video/graphics power just needs the manufacturers to supply drivers for */Linux and there’s no need for Wintel at all. Graphics is not a feature of Atom processors in a legacy PC, and the more power they consume in a mobile SoC just drains batteries sooner. In my whole life, I have never bought a video card because of its ratings. I always paid attention to price and video-resolution. About the only time I remember actually buying video cards they were PCI cards to create multi-seat systems for ~$40 each and similar cards for servers that would not boot without video cards installed. I had intended to install with a temporary card in place… Why a server sitting on a rack needs a video card, I never figured out but that BIOS had a mind of its own and a couple of other bugs too.

  42. Deaf Spy quoth, ” the performance of Atom on Windows 8 is greatly superior to Tegra 3 on Windows RT”

    Except that the Tegra loaded web-pages faster… I don’t agree with their conclusion, obviously. The task most people do most often is the thing we need optimized.

    Further, Intel can and does respond to competition but it had to go to what, 22nm to beat ARM at 28nm? That’s thanks to the burden of x86. It took Intel years to catch up in which time ARM has taken huge market-share. ARM can invade Intel’s turf at will while Intel is still years away from doing much in ARM’s territory. ARM has out flanked Intel which too long clung to Wintel as the only game in town. Then, there’s still that price-differential… Both M$ and Intel are going to have to cut prices. Both seem intent on raising prices on the top end to pay for that cutting but that only ensures that ARM will have a bigger slice of the pie eventually.

    Oh, and these benchmarks were done with that other OS. As we well know, M$ is not adverse to tweaking things depending on what else is in the system…

  43. oiaohm says:

    Deaf Spy arm enters and leaves power management many times faster than x86. So Roberts Workload many in fact be better serviced by arm. Why idling on an armv7 sees most of the silicon turn off.

    This is the big difference between armv7 and intel. In web serving where people want a decanted server yet there work load is not constant the the arm v7 can leave the intel for dead.

    Deaf Spy the benchmarks you are still looking at are armv7 not armv8. Armv8 even that it 64 bit has a lower power usage profile.

    Deaf Spy armv8 have been out for a while.
    http://www.xbitlabs.com/news/cpu/display/20121030140134_ARM_Introduces_First_ARMv8_Cores_Cortex_A53_and_Cortex_A57_Make_Debut.html

    Armv8 use 1/4 the power of a armv7 under full load.

    Armv7 turns on too much under heavy computation loads. Reason for big little in Armv7 chip-sets in phones and tablets. Hello bug. Armv8 the chips are designed segmented so even under complex loads the chip is never fully turned on. This is how come it pulls 1/3 of the 32 bit version at the same load level.

    Really intel has done a huge power opts to get there x86 chips so far. The big thing to keep up with arm intel has to keep cutting power or focus on providing cpu features arm does not have that do improve security.

    Also notice how much that showed that is not the CPU part. Its the gpu part.

    Deaf Spy yes a year or two is when the game between x86 and arm get dead serous arm and intel are still finding ways to cut power. Problem is both are finding radical power reductions.

  44. Deaf Spy says:

    Many thanks, Mr. Pogson. Now we’re speaking numbers, and I am pleased. The discussion in again on spot.

    However, I was saddened to find the following conclusions in the article you quoted.
    “Although these uber-granular numbers were generated in Intel’s lab, we can at least confirm that, using Chrome, Safari, or IE10 on Samsung’s ATIV Smart PC 500T, the performance of Atom on Windows 8 is greatly superior to Tegra 3 on Windows RT.”
    Now, let’s also consider the power consumption:
    “Now, with that in mind, the Atom-based Acer also delivers better power performance. Moreover, it’s clear that the complexity of a page being displayed affects consumption, Intel’s Atom demonstrating its strengths in the GPU and memory categories as Nvidia’s ARM-based cores use slightly less power on the more complex webpage. ”
    “The claim that x86 ISA suffers an inherent efficiency disadvantage to ARM does not hold true when you break down the power consumption of currently-available platforms sporting both architectures. They finish neck and neck in most cases.”
    “In general, our analysis suggests that the ARM-based CPU core is excellent at doing nothing, but starts to require considerably more power during computationally-intensive workloads. ”

    This is not good for your cause, Mr. Pogson. Looks like Atoms are more powerful, but still in reality consume as must overall power as ARMs. And now we have Atom Bay Clover, which is even better. Btw, I have on of them at home. Goes easily to 8 hours with single charge doing browsing with flash and HTML5 pages, Skype calls…

    Sign. I guess we need to postpone Intel’s demise with a couple of years at least.

  45. oiaohm says:

    1. Intel vs ARM performance
    2. Can ARM be “close enough” to intel to count as a desktop replacement?

    Ok I think you will understand why these are not that important. After I do the power users examples.
    1. Photo editing
    Lets take Photoshop for example. If you have a powerful GPU 90 percent of photoshops processing is done in the GPU. Now look at gimp and other open source editors same trend. Look at android opengl es used massively.

    So photo editing you are not talking about CPU bound operation.
    2. Video editing
    Again this is GPU and Accelerator cards.
    3. Sound editing
    This gets really odd. You would think a intel would be good here. It turns out in sound editing real time ish. The more cores you have the better. So that each stream can be decanted processed. So a 30 percent slower per core but with twice the number of cores is in fact better for sound editing.
    4. High end gaming
    Again this is another one that is GPU bound.

    So power user cases really have very little todo with the cpu performance itself. Power usage is more of a question if you will have the drivers for the GPUs.

    Even in the heavy server room data processing GPU and accelerator support is more important that the cpu.

    Most people don’t need more than a core2duo if they have a decent video card. A core2duo without a decent video card you are in trouble.

    So with arm vs intel the question is more if Nvidia and AMD will release there top end video cards to be used with Arm.

  46. ssorbom says:

    Edit: Intel and ARM can “only” be comparable if you can take an ARM pc/laptop and drop it in place of an Intel pc/laptop.

  47. ssorbom says:

    There seem to be 2 threads in this discusson:
    1. Intel vs ARM performance
    2. Can ARM be “close enough” to intel to count as a desktop replacement?
    Point 1 is just a matter of benchmarks.
    Point 2 can only be won in ARMs favor if you can take an ARM pc/laptop and drop it in place of an Intel workstation/laptop. For that ARM configuration would need to compare favorably with at least a Core2Duo.

    Here’s a problem I have with the current discussion:
    Point 2 is only being argued in terms of the “average” use case scenario. Some people seem to be claiming that anything requiring lots of power is a niche case. If I may be permitted to suggest it, lets zoom out a bit.

    There are four non-IT “Power usage” scenarios that I know of:
    1. Photo editing
    2. Video editing
    3. Sound editing
    4. High end gaming

    By themselves, each may be considered a niche case. However, taken as a whole they probably repressent a signifigant market share. They usually need power, and thus far, the only CPUs I have seen that meet said requirement are made by Intel and AMD.
    Even if they don’t repressent a majority market share, where is the justice in throwing these people under a bus?
    Smartphones will win the day only when they can do these things WITHOUT the cloud. Until then, the PC/laptop will continue to have a very loyal fanbase.
    regarding 100% CPU usage, I sometimes hit those limits too, and I don’t use Windows very much. I’m not talking about my server either.

    Also note, the usage scenarios I mentioned so far don’t include heavy number crunching, servers, or commercial Animation. I am speaking strictly of everyday “power usage”

  48. dougman says:

    Shedlock: Umm who cares, we are discussing the downfall of Wintel, Arm and the death of x86. Go play with your malware infested computer thats running at 100%.

  49. oiaohm says:

    Deadlock the latest Qualcomm you are referring is not the latest.

    There is no benchmarks to the Qualcomm’s Snapdragon 410 that was release in December to ODMs.

    There is a problem Armv8 is in the latest generation iphone and yes it has also showing a performance increase over prior Armv7.

    Amd will be releasing a 16 core armv8 chip. Goes into the same motherboards as Amd x86 server chips. Get more fun multi cpu boards can have a mix of Arm and x86 on the same motherboard. Really this is when we will be able to put a arm and x86 designs really under heavy testing.

    Amd x86 server cores stop at 8 cores. Notice something if there is a 30 percent performance difference will it really matter when you get double the number of cores with arm?

    Yes its not going to be quad core vs quad core. Less silicon space of arm cpus cores equals more cpu cores per chip.

    Really you are not talking are large enough performance difference for people looking arm to care. Yes intel need to be making arm look 50+ percent slower not 30 percent.

  50. Deadlock says:

    “One thing to note, Intel has taking notice of Arm by introducing Haswell. ”

    You are aware that both Haswell and the upcoming low-power “Bay Trail” (which beats the latest Qualcomm ARMs by 30%) are x86 and will quite happily run WIndows?

  51. dougman says:

    One thing to note, Intel has taking notice of Arm by introducing Haswell. The more recent Chromebooks are built around Haswell processors, how well Intel fares as a result is anyone’s guess.

    http://arstechnica.com/gadgets/2014/01/toshiba-throws-its-hat-in-the-ring-with-13-3-inch-haswell-chromebook/

    “I cannot shake the impression that the lion’s share of the comments I see complaining about poor specs are from people that either 1. play hardware intensive games; or 2. are victims or MS bloat. I for one am happy to see movement away from power hungry CPUs that can serve anybody (until the next MS bloatware comes out, anyway) to optimized Linux computers that serve 95% of the non-gaming population very well.”

    x86 is NOT required anymore all PC/Laptop OEMs have been loosing money in the last 5-10 years because of Intel and Microsoft’s control over the whole PC/Laptop industry’s profit margins, now with ARM Laptops, the industry can make significant profits and they can grow, their products can thus much improve, innovation can happen and that means better products sooner for everyone.

    I do know that HP is coming out with 64-bit Arm servers this year and that Ubuntu has been working in conjunction them in the process.

    http://www.ubuntu.com/partners/hp/moonshot

    So in Summary, bye bye Wintel.

  52. Deaf Spy wrote, ” Name one ARM CPU with the same or better CPU throughput than a x86 core.”

    Tom’s Hardware and Intel tested current x86 z2760 v ARM Qualcomm’s APQ8060A Dec 24, 2012 and web browsing showed pages loaded faster with ARM than Intel.

    Further, Tom found “Powered by a 2GHz dual-core Intel Atom processor, the smartphone scored more than 27,000 points on the AnTuTu benchmark. Comparatively, the HTC Droid DNA settles for a score of 14,000.” when testing a smartphone.

    However, another test showed Tegra4 and K1 beat that with 35000 points.

    So, slice and dice it any way you like. ARM is certainly competitive. Just ask consumers. The results that show Intel ahead are usually comparing Moore’s Law resolutions, not architectures. Certainly at a fine enough resolution x86 can best ARM but who cares now that both of them are producing cores that are tiny enough that folks are adding cores they mostly don’t need? ARM still wins on price.

  53. Deaf Spy says:

    “Blah-blah-blah, yackity schmackity”, if I may quote a favourite animation character of mine.

    Now, now, Mr. Pogson. Please spare me the philosophical exercise, where you throw stones in all directions, but one. I humbly urge you to focus on the question at hand. Name one ARM CPU with the same or better CPU throughput than a x86 core. CPU throughput. Not price, not power consumption, not TCO, not size. Just the CPU throughput. That thing you mentioned youself. How can I be trolling when these are exactly your words?

  54. Deaf Spy wrote, “Numbers, Mr. Pogson. Numbers, please.”

    Again, the trolls assert without evidence that my preference for one technology over another is irrational. It is not. I can calculate price/performance. I can calculate the total cost of ownership. I can weigh the environmental costs of the Wintel treadmill. I have paid the price for the “Intel” label many times. I have reasons for my preferences. I have given numbers which the troll demands are not sufficient. Too bad. Chuckle.

    Meanwhile, the production of small cheap computers outstrip anything imagined in the Wintel world. This year, in a single year, more non-Wintel system will be shipped than the total installed base of Wintel systems in existence. Bang! The Wintel bubble has burst. The explosion we heard in 2013 was just the sound of the match being struck. 2014 is the end of Wintel in so many ways: Wintel is no longer the default client OS or server OS and everyone has the choice of */Linux on ARM and they know it. Chuckle… 😉

  55. Deaf Spy says:

    Numbers, Mr. Pogson. Numbers, please.
    Numbers that demonstrate how Exynos 5 Octa has more CPU throughput than your Beast.

    As a physicist you should now well that in Mathematics, Physics and IT assessments such as “pretty snappy” are irrelevant.

    Btw, an 8 core Xeon swipes the floor with 16-core Opteron on both CPU throughput and virtualization.

  56. deadlock says:

    Beast idling is less throughput than most of these modern ARMed processors in smartphones going flat-out.

    There are too many. There’s Exynos 5 dual-core at 1.7gHz. That beats Beast idling.

    Please stop embarrassing yourself by comparing ARM at full load to x86 at idling.

    “Apples to apples”. Is it such a difficult concept?

    Beast’s CPU is AMD64 Phenom II at 2.5gHz.

    It absolutely smashes the octa-core Exynos 5 too. As does the original Phenom at 2.6Ghz. The octa-core Exynos just about manages to beat an AMD dual-core from 2006.

    YOU wrote – “Beast has 4 cores at 2.5gHz. There are ARMed CPUs today with that much throughput.”

    The Exynos 5 does not.

  57. ram says:

    ARM or x86 architecture, or perhaps even something else, it is not clear which one will “win” or how the market will divide up. No doubt lower $/MFLOP and Watts/MFLOP will be important for all of them.

    With respect to clusters: Intel, AMD, and NVidia (among others) all sell boards with large numbers of processors on the board. They can be used for general purpose high performance computing using OpenCL. All the offerings are pretty good, the benchmarks are close overall. Again, all different architectures but no clear winner. They all work brilliantly under Linux!

  58. Deadlock wrote, ” Why are you so reluctant to simply name one?”

    There are too many. There’s Exynos 5 dual-core at 1.7gHz. That beats Beast idling. Exynos 5 was shipped early in 2013. Later chips put that one to shame, like Exynos 5 Octa which has 8 cores all pretty snappy. That’s just one manufacture and the last year’s news. There are a bunch of other manufacturers with sparkling devices.

    Beast’s CPU is AMD64 Phenom II at 2.5gHz. Full-out it is very useful but despite a ton of applications at my finger-tips it rarely approaches 100% utilization. Idling as it is most of the time, many ARMed CPUs could do more. This kind of throughput is not about production but responsiveness. Beast and these others are great multitaskers and can make ordinary people happy. That’s their purpose and they do it well at reasonable cost. Beast’s CPU cost me ~$200 about 6 years ago. Today, if I were building a new Beast, I could certainly use one of these ARMed system, not for the mobility, but for the throughput. I already have a tiny single-core smartphone which does what I need to away from my chair.
    PS: After posting this comment, I checked “top”. Beast was at 6% utilization with 0.15 tasks in the queue while I browse the web, type the comment, listen to music and have the world at my fingertips.

    Here’s a snapshot of what Beast does. The long pause was me fixing things in the garage after the “little woman” ran over a box… Beast occasionally gets something to do, does it in a hurry and goes back to idle. That’s IT since the GUI. Spikes might reach 30% briefly but they only last a blink or two. It’s point, click and gawk… This is why I can run 50 people on a single GNU/Linux box for peanuts.

  59. Deadlock says:

    Beast idling is less throughput than most of these modern ARMed processors in smartphones going flat-out.

    Compare apples to apples, please.

    That’s silly. There are so many

    If there are “so many” then finding one that’s equivalent to the quad-core in your server in benchmarks should be EASY. Why are you so reluctant to simply name one?

  60. DeafSpy wrote, ” Just name the chip, or admit you’re exaggerating bit time.”

    That’s silly. There are so many. Name any quad-core or more ARMed chip made for smartphones/tablets in the last year or so. They all run at more than 1gHz with some up to 2.5gHz. Poor old beast is quad-core and runs at 2.5gHz but rarely is busy despite my huge collection of local and web-applications.

  61. Deaf Spy says:

    Well, Mr. Pogson, and which ARM chip has the same throughput than a x86 chip?

    It doesn’t matter if your home CPU is idling. Just name the chip, or admit you’re exaggerating bit time.

  62. oiaohm says:

    Deadlock from what you stated you should be more up on arm benchmarks.

    Also you have to take into account that the snapdragon chip is 1/8 the price of the intel chip. So 1/3 performance is nothing. You can have more of them for the same money. The snapdragon arm cores use about 1/4 of the silicon each x86 cores do.

    http://arstechnica.com/information-technology/2013/06/amd-announces-its-first-64-bit-8-and-16-core-arm-based

    Yes 64 bit arm with 16 cores is coming.

  63. Deadlock demanded “Equivalent CPU throughput”.

    Beast idling is less throughput than most of these modern ARMed processors in smartphones going flat-out.
    Here’s Beast idling:
    “%Cpu0 : 0.0 us, 0.7 sy, 0.0 ni, 99.3 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
    %Cpu1 : 1.6 us, 3.5 sy, 0.0 ni, 92.7 id, 2.2 wa, 0.0 hi, 0.0 si, 0.0 st
    %Cpu2 : 2.5 us, 3.8 sy, 0.0 ni, 93.1 id, 0.6 wa, 0.0 hi, 0.0 si, 0.0 st
    %Cpu3 : 3.4 us, 4.7 sy, 0.0 ni, 91.9 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st

    For round numbers, I browsing, typing and clicking draw less than 10% utilization, like 0.4 cores working hard, similar to what these ARMed thingies can do. For the greater certainty, I have 180+ processes queued up but most of them do nothing until I click on something. I have two browser windows open with a bunch of tabs in each and windows to several other applications and a virtual machine. I have two databases and a web-server running too. Should I keep a Formula-1 racer in my garage in case I need to get somewhere faster? I don’t think so. That would be a very inefficient waste. So’s some big CPU idling most of the time.

  64. ssorbom says:

    I was looking at the Chinese Yeelong computers with MIPS architecture today, anyone know how they compare to ARM or Intel?

  65. ssorbom says:

    Ram,
    How many CPU’s can you fit on a single motherboard? Clusters are good, but kind of bulky right?

  66. ram says:

    Phoronix has had some articles on ARM and ARM clusters in benchmark comparisons against other architectures. A cluster of ARMs can do anything a high end Intel processor can do.

    It is all starting to come down to how many MFLOPS/Watt and MFLOPS/$ you get. It is becoming a close run thing that will probably squeeze out AMD.

  67. dougman says:

    “I’ll ignore the trolls on your side”….

    Deadlock, does this mean you will *ignore* yourself? By all means sir, please do.

    You came in here calling one person a shill and then outright demanding action on the subject of Arm, in a effort to make the site owner look ignorant.

    Actions speak louder then words.

  68. ssorbom says:

    Ohiahm,
    I read the article you linked to, and it seems that the benchmarks are only comparing Intel Atoms to the ARM. Correct me if I am wrong, but it still seems as though an equivalent ARM replacement to Intel PC chips is a bit of a ways off yet.

    On the other hand, if the ARM chips can run Word processors, IDEs etc decently I won’t complain. But my other objections still stand. Even on laptops, the OS seems to be ROM only.

  69. Deadlock says:

    I’ll ignore the trolls on your side, Pogson. They’re not worth a response, anyway.

    Deadlock also demanded that I “NAME ONE” CPU with ARM that can give Beast a run for the money.

    Some nonsense about smartphones, and trying to change the wording of the request to move the goalposts. No-one said “run for it’s money”.

    YOU wrote – “Beast has 4 cores at 2.5gHz. There are ARMed CPUs today with that much throughput.”

    You have so far failed to name one. You made a factual claim and you now seem to be trying very hard to avoid proving it.

    Here’s a smartphone using 8 cores at 1.5gHz. Beast might have better networking and storage but can’t beat that camera…

    Who cares a whit about the camera? I’m talking about the CPU. The processor. The bit that does the “thinking”.

    Stop trying to move the goalposts and waving your hands. It’s a simple enough request – name an ARM CPU with equivalent throughput to the quad-core in your server. Not “give it a run for it’s money”. Not with a better GPU. *Equivalent CPU throughput*.

  70. matchrocket says:

    New year, new trolls.

  71. dougman says:

    Deadlock has his computers running at 100% due to the fact its full of malware. Pretty bad that malware can fool Windows and spawn additional wuaueng.dll and svchost.exe processes.

    Also its pretty bad that Surface requires an i5, when I have a i3 in my NAS and it runs at <5% even when checking parity.

    But getting back to the storyline, I do not see X86 lasting past 10-years if that and Xeons will stick around for a good while make for better substitutes of over i7s.

    ARM also has an advantage on price. A few years ago, before significant numbers of smartphones were being sold, ARM said that a billion ARM chips were being made every year and the cheapest were being sold in quantity for 50 cents each.

    Currently there is (23) Linux distros built for ARM.

    Data-centers in reach for a better PUE, are always looking for ways to save energy. Liquid immersion, 380Vdc power and ARM based servers.

    http://www.droid-life.com/2014/01/05/nvidia-announces-tegra-k1-the-worlds-first-192-core-super-chip/

    http://www.theregister.co.uk/2013/12/16/google_intel_arm_analysis/

  72. Deadlock wrote, ” I have machines where running at 100% load for days is normal. And thirty percent faster means work is done 30% quicker, so yes, it’s really quite significant.”

    That is not typical of desktop users. The original point was about small cheap computers with ARM not being able to do what typical users want.

    Deadlock also demanded that I “NAME ONE” CPU with ARM that can give Beast a run for the money.

    Well, Nvidia Tegra has this monster with 4 general purpose cores, 1 for idling, and 72 for graphics. Beast has some old Radeon 2100.

    Here’s a smartphone using 8 cores at 1.5gHz. Beast might have better networking and storage but can’t beat that camera…

    Galaxy S4 smartphones are also pretty wonderful:

  73. Deadlock says:

    Relevant benchmark: “The first benchmarks of Intel’s upcoming Bay Trail SoC have appeared online — and it’s good news for x86 fans, but terrible news for ARM: Bay Trail-T, clocked at just 1.1GHz, is around 30% faster than Qualcomm’s Snapdragon 800 clocked at 2.3GHz, the fastest ARM chip on the market.”

    “Is 30% significant when CPUs are idling all day long?”

    Your usage again. I have machines where running at 100% load for days is normal. And thirty percent faster means work is done 30% quicker, so yes, it’s really quite significant.

    “Single core ARM CPUs were able to do 1080P video years ago. What more do you want a smartphone to do?”

    This is about comparative performance of CPUs, not performance of smartphones. Stop trying to move the goalposts.

    You’ve supplied a benchmark that flat-out disproves your ridiculous claim, then you hand-wave to try and make it irrelevant.

    You wrote – “Beast has 4 cores at 2.5gHz. There are ARMed CPUs today with that much throughput.”

    NAME ONE.

  74. dougman says:

    Apologist Deadlock, rushes in to defend MicroSh1t. Hoooorayyy!!!

    *Rolls-eyes*

    “So it’s not actually confirmed that it’s Windows at fault” Its not?… given all the pieces of the puzzle below, one begins to realize that Windows was the underlying fault. Its not hard to drop malware laden thumb-drives is it? Oh whats this thing, oh hey its a USB stick lets plug it in and see what it has on it. BANG!

    “A Target spokeswoman declined to provide any details on how the security breach occurred, but a variety of security experts agreed that it likely involved Target’s point-of-sale system, the software the company uses to carry out transactions at the cash register.”

    The malware used in the attack is called ‘Dexter’ and it was specifically designed for Windows.

    http://www.microsoft.com/casestudies/Case_Study_Detail.aspx?CaseStudyID=4000009407

    Malware Targeting Point of Sale Systems – http://www.us-cert.gov/ncas/alerts/TA14-002A

    http://www.seculert.com/blog/2012/12/dexter-draining-blood-out-of-point-of-sales.html

    This US-Cert bulletin from January 2nd mentions:

    “For quite some time, cyber criminals have been targeting consumer data entered in PoS systems. In some circumstances, criminals attach a physical device to the PoS system to collect card data. In other cases, cyber criminals deliver malware which acquires card data as it passes through a PoS system.”

    Dexter steals the process list from the infected computer, and dissects memory dumps looking for the track one and two data I mentioned earlier. At a certain point, the infected machine sends the captured data to the attackers’ command and control server. After which the criminals are free to use the information to clone new cards. The unfortunate thing is that as of yet, no one understands how the malware makes its way into the PoS system. Hint: Thumb-drives, inside help or directed social-engineering attack.

    http://www.infosecurity-magazine.com/view/36013/dexter-pos-malware-returns-to-target-holiday-shoppers/

    Consider that the majority of POS systems run Windows, I say this is a MiscroSh1t failure, but hey according the the official EULA MicroSh1t cannot be liable or held responsible, nor can they be taken to court with a class-action.

    Really makes you feel warm-n-fuzzy doesn’t it?

  75. oiaohm says:

    Deadlock low end x86 chips don’t beat 32 bit arm chips in ever benchmark. That is the problem.

    Deadlock arm chips can be packed in at a far higher density than a Quad core AMD chip.

    You also picked out a solo benchmark please go and look at the benchmark site I refered to. It was a broad compare yes the results are not black and white.

    Depending on your work load decides if arm or x86 is better. Yes it will be a large margin. Some workloads an x86 is 1/3 of the performance of a arm chip as well.

  76. Deadlock wrote, ” The latest ARM CPUs are still beaten handily by low-end x86 chips.”

    Low-end x86 chips include 1.6gHz Atoms. I have a couple. Clockspeed does relate to throughput. Both ARM and x86 use pipelining so that a couple of instructions are done each clock-tick. The most common operations are put this there. There’s not a lot of difference between CPUs on that except the number of registers. With today’s software-bloat, no cache eliminates the memory-bottleneck. So low-end Atoms and ARMed CPUs do comparable work, and thanks to Moore’s Law, do with comparable expenditure of energy, but Atoms still cost way more than ARMed CPUs so price/performance is for ARM.

    Deadlock also wrote, ” if ARM is so fast, why do you bother to offload tasks to a server”.

    That’s obvious: efficiency. If all your data, storage and RAM is in one place it will be better utilized than if scattered all over the network. e.g. in some schools where I worked, the lab might have 24 P4ish PCs with 512 MB RAM and 40 gB storage. Of course with thick clients each student gets to use only the one feeble CPU, 512 MB RAM and 40gB storage. By moving most of those resources to a powerful server, each student gets to use one powerful CPU, 12gB RAM and 1TB of storage. The possibilities of what can be done with such IT is limited mostly by imagination. One can also reduce power consumption of clients and save heat load to the room. Everything is better this way except full-screen video, which, in a room, can be handled with a video projector quite nicely for about the cost of one more PC. Then when you consider that storage can be faster and files cached in RAM, the better configuration blows away the competition. I have done similar many times and it shocks the users who were used to 30s logins. One can easily cache the web-traffic as well. There is no downside.

    Relevant benchmark: “The first benchmarks of Intel’s upcoming Bay Trail SoC have appeared online — and it’s good news for x86 fans, but terrible news for ARM: Bay Trail-T, clocked at just 1.1GHz, is around 30% faster than Qualcomm’s Snapdragon 800 clocked at 2.3GHz, the fastest ARM chip on the market.”

    Is 30% significant when CPUs are idling all day long? I don’t think so. Now, about prices… Is a quaterback in fear of his job because a lineman outweighs him by 30%? Nope. Single core ARM CPUs were able to do 1080P video years ago. What more do you want a smartphone to do? A few more processes? OK. Then what? ARM never needed to beat Intel on raw speed just price/performance. Monopolies hate competing on price/performance. There’s a reason for that.

  77. Deadlock says:

    @Pogson

    “Nonsense. I use such all day long and until I build a kernel, Beast rarely gets up to speed. CPU load-meter shows 2%. If Beast ever gets busy, it’s an I/O bottleneck most likely.”

    How does that work? CPU is starved of data to work on because I/O (disk or network) is not fast enough, so CPU gets busier??

    “Beast has 4 cores at 2.5gHz.”

    x86-64 cores, not ARM. Compare apples to apples, please.

    “There are ARMed CPUs today with that much throughput.”

    Name one! Name just ONE ARM CPU with the throughput of a quad-core server-class AMD CPU. The latest ARM CPUs are still beaten handily by low-end x86 chips. MIPS is not an indicator of the actual work done by the processor, so don’t even bother to bring up equivalent MIPS numbers. Benchmarks or GTFO. And if ARM is so fast, why do you bother to offload tasks to a server?

    ” There may always be some need of a desktop box but it does not need to be a hair-drier. Notebooks proved that. Netbooks proved that. Now smartphones prove that. It’s not about RAM or CPU power, it’s about storage. Folks don’t need so much local space for storage because storage and network storage are so much more capable than before. I used to use ATX boxes with 10 HD slots. Those would be silly these days of TB storage. We no longer need 400W PSUs to run these smaller systems which are much more powerful than anything needed for typing/browsing/playing multimedia. People are being productive all day long with little more than a smartphone. They can produce audio, video and stills by the gB. They can store them on the server or in the cloud. It’s all good.”

    Translation – “My usage scenario is more important than that you use a computer for. My narrow and limited experience trumps your actual needs.”

    @Mats

    “blamed the breach on either the security of Target’s Windows-based Point-of-Sale systems or the company’s failure to fulfill its security obligations under the Payment Card Industry Data Security Standard (PCI DSS).” [emphasis mine]

    “Thanks again Microsoft for your great software.”

    So it’s not actually confirmed that it’s Windows at fault, and could be (and is most likely) a failure of Target to follow the policies and procedures of a *standard*. Yet you automatically blame Microsoft?

    When does your shill check arrive from [Canonical/RedHat/Debian]? Do they pay per biased conclusion or made-up ‘fact’ about Microsoft, or per-post?

  78. oiaohm says:

    ssorbom libreoffice beta and openoffice ports work fairly well on android.

    Andlinux options also run at quite decent speeds on android.

    arm vs x86 get very complex. Differences in instruction set really do make sure you cannot do Ghz to Ghz that well.

    http://www.notebookcheck.net/SoC-Shootout-x86-vs-ARM.99496.0.html

    The big thing is performance per watt arm still wins all that all the time.

    Note we are not into arm64 yet. Arm64 bit chip design is faster at the same clock-speeds.

    Data centres are looking to deploy more and more arm. The low power usage has the attention.

  79. Mats Hagglund says:

    More about Target and the fiasco – the background:

    http://www.microsoft.com/casestudies/Case_Study_Detail.aspx?CaseStudyID=4000009407

    In 2004:

    In 2004, Target joined the Microsoft Technology Adoption Program (TAP) for virtualization and found the solution it was looking for. During the TAP, the Microsoft team worked closely with Target Technology Services team members to virtualize the Linux-based pharmacy solution and run it successfully in a Microsoft Virtual Server 2005 environment. “Microsoft was pretty creative about developing a virtual machine edition for SUSE Linux for us,” says DeBrine. “So the first virtual instance in our stores was actually a Linux virtual machine, and it worked great. We deployed that solution to approximately 1,500 stores.”

    Later:

    Target’s participation in the TAP program absolved the company from having to buy new hardware for its pharmacy applications and was the catalyst for Target choosing a Microsoft Virtualization solution for every one of its retail stores in the United States. However, Target Technology Services did due diligence in investigating alternative solutions. “We looked at VMware ESX Server in 2005, and part of the conversation was around the price of deployment,” says DeBrine. “Instead, Microsoft gave us everything we needed in a virtualization solution at lower cost. Given the pharmacy server experience, we knew Microsoft would be there for us, so we chose a Microsoft Virtualization solution as a key cost-optimization strategy for Target moving forward.”

    —————

    Perhaps Target is now happy?

  80. ssorbom says:

    It is available in google playstore. I checked the last time Ohiaohm recommended I go there.
    https://play.google.com/store/apps/details?id=com.andropenoffice&hl=en
    You might be thinking of libre office. I know they don’t have a port yet.

    This is promising news then.
    Some people give docking a bad rap, but I really like the one that integrates with my Dell D620 laptop. They are decent if they plug straight into the motherboard. I’d be curious to see what your experience of using OpenOffice directly on Android is and how it compares to standard Open Office.

  81. Mats Hagglund says:

    70 million more reasons not to use Windows:
    http://www.computerworld.com/s/article/9245335/Breach_goes_from_bad_to_worse_for_Target_i_and_i_its_customers

    And you know what, they once again didn’t mention why. So let’s see why.

    “The new revelation does represent a new breach, however, or at least the breach of an unrelated system during the period covered during the same attack, according to the few details Target has released. Most analysts and news outlets have blamed the breach on either the security of Target’s Windows-based Point-of-Sale systems or the company’s failure to fulfill its security obligations under the Payment Card Industry Data Security Standard (PCI DSS).”

    http://yro.slashdot.org/story/14/01/10/2321209/target-admits-data-breach-may-have-up-to-110-million-victims?utm_source=rss1.0mainlinkanon&utm_medium=feed

    Thanks again Microsoft for your great software.

  82. ssorbom wrote, “How well does Open Office run on Android for instance? I only have a dumbphone, so I can’t test it. Is it decent?”

    I don’t think it’s ported yet. OpenOffice.org is designed for a resourceful computer. Web-apps designed for editing work very well. I use one on this blog, for instance, TinyMCE. OpenOffice.org can run on GNU/Linux on a smartphone very well.

    e.g. see http://www.youtube.com/watch?v=wzc0uMXGFBY

    That’s a video from 2012 of a smartphone running Ubuntu GNU/Linux and Android/Linux simultaneously with the ability to read and edit LibreOffice docs. That’s dual-core. Imagine what one of the newer huge-RAM-multiple-CPU solutions can do. The smartphone is no problem at all. The display, keyboard and mouse are key. Dock the smartphone to a monitor/keyboard/mouse and it is a full-fledged PC.

  83. ssorbom says:

    My knowledge of how processors work is kind of sketchy, but my understanding is that 2GHz ARM is not equivelent to 2GHz Intel chips, due to the use of a totally different instruction set. Even within Intels own line, I was told that a Xeon processor and a Core2Duo perform different tasks with different levels of efficiency.

    You touched on another sticking point I have with these machines too.
    Simply put:
    I don’t trust public clouds. I might use them for things I intended to share anyway, but never EVER for private stuff. It is my understanding that a lot of phone apps rely on the cloud to make up for shortcomings in power/space. When I talked about running apps below, I was referring strictly to whatever can be run with zero internet connectivity. How well does Open Office run on Android for instance? I only have a dumbphone, so I can’t test it. Is it decent?

  84. rudregues wrote, ” These software lock-in people in a manner it’s impossible to stop using Windows Desktops for them right now”.

    Munich showed that very few of those applications are “necessary” and none of them are needed by everyone. You do remember that people went to Moon before AutoCAD? If AutoCAD were so essential they would sell a lot more licences than they do. Locked-in friends can be freed.

    We have seen repeatedly that governments and large organizations feeling locked-in have been able to free themselves. In some cases they may have to write new software. It can be less expensive than paying some tyrant forever. In other cases, they can integrate several applications transparently. With X and FLOSS that’s pretty easy. Pick any application and you can match some of the most-used functions with FLOSS packages quickly. Computer science is not that complex. Pick any task and break it down into the component steps and software falls together rather quickly. I once replaced 6 man-years of software with ~1000 lines of code using a good programming language for the task. The new programme worked better because I had used the old one and hated it passionately. Businesses and such can pay programmers instead of licensing fees and they don’t need to write all the code just key parts of it if they use FLOSS.

  85. ssorbom wrote, “PCs will still have a (very large) niche…anybody actually USING productivity apps probably needs a more powereful processor than ARM — at least for now”

    Nonsense. I use such all day long and until I build a kernel, Beast rarely gets up to speed. CPU load-meter shows 2%. If Beast ever gets busy, it’s an I/O bottleneck most likely. Beast has 4 cores at 2.5gHz. There are ARMed CPUs today with that much throughput. There may always be some need of a desktop box but it does not need to be a hair-drier. Notebooks proved that. Netbooks proved that. Now smartphones prove that. It’s not about RAM or CPU power, it’s about storage. Folks don’t need so much local space for storage because storage and network storage are so much more capable than before. I used to use ATX boxes with 10 HD slots. Those would be silly these days of TB storage. We no longer need 400W PSUs to run these smaller systems which are much more powerful than anything needed for typing/browsing/playing multimedia. People are being productive all day long with little more than a smartphone. They can produce audio, video and stills by the gB. They can store them on the server or in the cloud. It’s all good.

  86. ssorbom says:

    Hang on,
    “Not growing” != “Dead”
    This is kind of touching on something I mentioned earlier. PCs will still have a (very large) niche. I was impressed the last time when Ohiaohm showed me the compiler link in the googlePlay store, but anybody actually USING productivity apps probably needs a more powereful processor than ARM — at least for now. Again that may change a few years hence, but were tot there yet.

    Also with regard to the repeated predictions of the “death” of the pc,:
    Regardless of what economics people want you to beleive, a saturated market is not a dead one. people will always need replacement parts for their existing pc’s (or whole new boxes), at least until ARM chips catch up. Intel can probably show enough return to continue making PC chips for a while yet. Also consider that a large portion of the PC market comes from the bussiness sector. Even if smartphones were able to provide a drop-in replacement for PCs by tomorrow, bussinesses would be slow to change toward the new paradigm. Not because the old way is inherently better, but because they have too much invested in the previous generation’s technology to dump it overnight.
    Finally there are also some hardware limitations on smartphone technology that really piss me of:
    1 All the peripherals I have seen for smartphones are horrid. Touch screens? Yuck. Any attempt to graft a decent keyboard onto existing smartphones is either prohibitively expensive (light based keyboards) or too bulky, thus cutting portability out of the equation.

    2. OS design/integration:
    A relative once told me that in the smarthone world, OSs are essentially treated like BIOSs, making changing underlying software a potentially dangerous task. When I looked into installing Duokan on my Kindle 3 (an alternate kindle OS) I was warned that one wrong move could brick my system. Whats up with that??

  87. rudregues says:

    Mats Hagglund:

    remember specific programs that are M$ only (Revit, Autocad etc). These software lock-in people in a manner it’s impossible to stop using Windows Desktops for them right now. Your point is valid for general people and I can see it as my friends stopped acquiring legacy pc’s, except for locked-in friends…

  88. dougman says:

    M$ is a thing of the past. “Just as the stone ax gave way to one made of bronze, the Windows PC must now give way to better, more customized, more refined tools.”

  89. Mats Hagglund says:

    They are dreaming of a pc boom near in future. But it looks like many, perhaps majority have already bought their last pc, at least Windows desktop/laptops. One thing easily forgotten is the demographics: markets are moving more to China-India-Africa-Arab world- Brazil, to the south. It means more mobiles, more very cheap but decent thinclients.

Leave a Reply