Oh Boy! Tilera Servers Out in March

“the ramp for the Tilera chips has been pretty steep, with over 80 engagements with system and network equipment vendors of all colors and stripes, and 20 design wins where the company has committed to use a Tile processor”

Wow! A Tile processor uses a bunch of RISC CPUs on a chip in a mesh. They have 64bit processing and 40bit addressing. The idea is to get close to one processor per thread so that fewer context switches and massive parallelism will get a lot of throughput at lower cost than x86 with SMP. For servers this makes a lot of sense and because they are optimized for Linux and have tools, porting is trivial. Lots of software that runs on GNU/Linux will be able to move quickly to servers running these things. Sampling is happening and production will happen in March. 2012 will be even more interesting than Android/Linux v world.

Of course, there will be particular applications where Tilera offers no advantage but there are lots that will. Anything on a busy website should be helped. Super-computing is a natural. Big data too… I would bet that as this matures, Google and other big players will slurp them up. That could really dent the bottom line of Intel/AMD which means price/performance could improve a lot for us ordinary mortals who do one thing at a time.

I would not be surprised if this technology eventually gets to the desktop. Most desktops don’t need this but terminal servers may be natural. Many of us have ~100 processes running. A hundred users might benefit greatly having one processor per process.

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. Bookmark the permalink.

47 Responses to Oh Boy! Tilera Servers Out in March

  1. falde says:

    “Change will not even be Considered until the vendors of that software offer product on ARM under an ARM version of Red Hat, which must in turn be offered on a Main stream platform from the likes of HP or IBM.”

    Are you saying that you will not plan ahead for this change but just do it when the product is there?

    That does not seam very “enterprise” to me. Also an enterprise that only use a single software product? You only run an ERP and nothing else? And furthermore totally unable to switch to another ERP software provider that gives more ROI?

    Mainstream hardware providers already have ARM-based hardware. Red Hat is working on ARM-support, currently in the Fedora branch. We are talking about the future here, not about what is available here.

    Also just because you have your head in your ass, do not assume that everyone that think different from you is not working with enterprise system. That is plain rude.

    Wine also are perfect for enterprise systems only a complete idiot (at least when it comes to wine) will think otherwise. It runs all of our legacy systems flawlessly on RHEL servers. That includes Win16 applications on our 64bit processors, as Wine does 8086 emulation.

    I have not heard of a single case where the CodeWeavers enterprise support team has failed to deliver flawless support of legacy enterprise applications. I have heard of plenty of cases where migration to Java is less expensive but that is a whole other issue. If you have a case where the CodeWeaver have failed to deliver, then please amuse us with evidence for this.

    In fact I believe that in some cases binary translation integrated with Wine is the best way to allow windows x86 application to run as smoothly as possible in a enterprise environment. That is for 32/64-bit applications. For DOS or Win16(requires DOS) binary translation is overkill, the 8086 emulator in wine will do the job.

    But of course, in many cases full migration is more cost efficient and also allows for usability redesign and added features.

    Running a full operating system like Windows with binary translation is just plain waste of computing cycles. Abstract emulation/compatibility layers should be native to the target, and binary translation performed only for the applications and libraries that are not part of the compatibility layer.

  2. oldman says:

    “oldman those parameters you are operating to are based on a presume there will be no arch changed.”

    And when they change they will change. Until then I am operating on them, and no amount of walloftext on your part is going to change that fact. Nor is it going to change the fact that all of my ERP systems are built on closed source proprietary. Change will not even be Considered until the vendors of that software offer product on ARM under an ARM version of Red Hat, which must in turn be offered on a Main stream platform from the likes of HP or IBM.

    If EVERYTHING goes as you think it will we are looking at 5 years before the “new” technology is at the point where we can begin evaluating it and probably another 3-6 years beyond that until the transition is complete.

    Hence my comment taht I would be retiring before any such change took place

    For as you should well know sir, enterprise IT doesn’t just hack things together as you seem to like doing.

  3. oiaohm says:

    oldman those parameters you are operating to are based on a presume there will be no arch changed.

    This is where the problem comes in. You and I have both seen the the disruptions an arch change will cause. Early adopter was amazon this happened 2 years ago.

    This is the problem. We are now moving from Early adopter to massive deployment.

  4. oldman says:

    “Oldman nothing you have said changes the fact the writing is on the wall informing everyone that x86 is ending.”

    And nothing you have said changes that facts of the parameters that I operate in. Regardless of what happens, there is an information life cycle that I operate in that will not be circumvented.

    Attempting to lecture some one in enterprise IT on the evaluation of a technology before it is even in the early adopter stage shows a level of misunderstanding of the realities of a real enterprise. I question sir whether you have been even within a stones throw of enterprise IT.

    If you are truly in a position to effect future planning for an enterprise IT shop, then I suggest you put your money where your mouth is and do so there.

    And spare me the wall of text on irrelevancies or your inevitable dismissal of me as an idiot. You know full well that I could care less what you think.

  5. oiaohm says:

    oldman the problem is even with VMWare and other hyper-visor for bigger data centres larger than you 13 physical hosts you are talking about.

    Hyper-visor not longer cutting. Hyper-visors dug the large data centres out of there last big power and space crunch. From big data-centres the tech trickled down to you. You lose upstream you lose your new great tech to upgrade to.

    All the advantage of Hyper-visors in big data centres is tapped out.

    You can only virtual machine so far before you hit max capacity. Some of these data centres are at scary level max capacity. Where a full multi nuke reactor setup is required as power source. Facebook alone takes enough power to power a small EU developed country. Most of England in fact could be powered by how much Facebook uses. Google sucks down even more.

    If we keep on the x86 path we are going to get to the horrid point where the most power consumed on earth is consumed by data centres.

    oldman the simple fact of the matter high end is going arm or mips or something else that is way more power effective than x86. Taking out planning permits to build nuclear reactors is not simple. Changing arch from x86 to arm or mips or something else is simpler might be painful but kinda no other option. This can buy them another 10 to 15 years before they have to upgrade there data centre power feeds. It also increases capacity of data centre through roof. 10 to 15 with full data centre arm compared to 0 with x86 and its next generation. Next x86 chips use more power for space.

    High end has basically at long last hit a solid wall for future x86 usage.

    The problem is what makes the hardware you use cost effective oldman is the volume the high end buys. No high end volume buys on x86 servers and high end buying arm will see the x86 server price go up. This was seen with powerpc and sparc chip servers. When they were selling high end they were cheaper.

    This is not a if but or maybe. So arm in server room is a given. Its going to a issue of HP and other arm makers will be able to keep up with the demand from the high end server rooms for arm and mips and other low power solutions. Since its not a option for a lot of there server farms to upgrade there x86 chips without at least moving some work load to arm to reduce power usage.

    At most there is only 1 to 2 more x86 generations upgrades that will be performable in the large data centres. That is only possible if a percentage of the data centre converts to arm or mips.

    oldman the point is you have years. This gives you time to address any problems. So they should not sneak up on you. Waiting until is paramount is bad planning.

    The price changes in stuff should start happening over the next 12 to 24 months. As the high end moves over to more arm and mips and other techs.

    Next thing without the high end servers driving x86 the performance of x86 chips compared to everything else will drop off.

    There is no magic software solution this time. Hardware is what has to be changed this time.

    Necessity will come from cost and performance oldman.

    I have been watching datacentres trying to find more and more ways to stack more and more servers onto the same poor x86 chips. Its like the straw that broke the camels back. At some point the camel cannot lift the load. That is where we are the camel that is x86 can no longer lift the load. We need a new better bread of camel. Arm and mips are both better breads. Even tile chips are better breads limited licensing of tile makes it not so popular.

    The writing is on the wall oldman. The only thing the writing on the wall is not telling us is exactly when. There is no writing on the wall suggesting a if x86 does something it will get back in the game.

    The technical limit of x86 dia size has it check mated. AMD in some of there recent processes tried a desperate design move to try to get around the dia limit. This lead to the under performance x86 chips AMD made. If that had of worked. That would have given x86 a few more years.

    Next desperation move is taking out x86 cores and replacing them with gpu cores on the dias. Again it desperation. Gpu cores cannot allocate memory directly so bottle neck in particular usages. Also arm chips can also have gpu cores on chip. So you are still losing the power game.

    x86 instruction set no longer cuts it. There is only a few years that desperation moves can be done.

    Oldman nothing you have said changes the fact the writing is on the wall informing everyone that x86 is ending.

  6. oldman says:

    “It could be a year or two off yet but a memo might come down from On High to save energy no matter what.”

    That is all fine and Good POg, but the reality is we are already in the process of implementing energy saving without having to do a forklift upgrades of current technology. Our server virtualizsation efforts using VMWare virtualization have already resulted in huge savings in avoided costs, and we are in the process of formally moving to to a virtualize first policy that has allowed us to collapse 350+ servers into 13 physical hosts. x86 Virtualization works and works well, and does not require massive dislocation.

    ARM technology may indeed make it into the server room, but with efforts like ours in play in IT at large, it is my belief that it will be years before the necessity to move becomes paramount.

  7. It could be a year or two off yet but a memo might come down from On High to save energy no matter what. There are buildings which have maxed-out their electrical distribution system and need to look at better technology. The worst I have seen was 7 servers in a room designed for none and 24 PCs in a room with no special ventilation. At -40C outside we were sweating in there. Moore’s Law and/or ARM will save us whichever comes first. I bet on ARM just because of the architecture. For the same size of transistor, ARM uses less power.

  8. oldman says:

    “You have time at the moment oldman. Maybe you want to wait until moonshot and other items start replacing server farms. That is basically 12 months off. But that is 12 months of prep time lost that you will not be able to get back.”

    I’ll tell you what Mr oiaohm. If a year from now I am Beginning the planning for a mass migration to HP ARM servers into my shop while I am learning the ropes on my new ARM portable. You and Pog will be the first to know.

    Of course that will be just after the mass sightings of flocks of flying pigs at the start of the Palin administration in the US.

  9. oiaohm says:

    oldman really you don’t pay with next generation tech.

    What I am suggesting is not exploring technology its keeping a heads up on what is going on.

    MS is demoing Windows 8 Arm devices. There are formal announcements of plans.

    What I have described is happening. Arm is becoming the common processor type.

    Of course a person like you oldman is not part of Project Moonshot. You are too far down the food chain.

    Simple fact of the matter oldman you have your head in sand. x86 has been beaten. The effects are going to take a few years most likely to get to where you are oldman. I am looking at a year before it starts effecting me in a big way.

    So yes it start locating your no longer maintained applications. They are going to bite sometime in the future. You have time to plan migration to something that is currently maintained or make replacements.

    Now if you don’t start now by the time everything catches up with you. Most likely you will have too many problems.

    You have time at the moment oldman. Maybe you want to wait until moonshot and other items start replacing server farms. That is basically 12 months off. But that is 12 months of prep time lost that you will not be able to get back.

    Secuirty alone getting rid of anything from vendors that is not maintained is a good thing.

  10. oldman says:

    “oldman maybe I am management the guy planning the future live cycles.”

    And Pigs have wings…

  11. oldman says:

    “Those requires you can bet management will force on the new Arm laptops. That you will have to solve. Basically this is just an unwise excuse. Justification why you don’t have to be ready oldman.”

    What exactly am I getting ready for sir. A potential market for ARM laptop running an non existent OS running non existent applications?

    Give me a break!

    “I love oldman my management is at fault not me. So I cannot be blamed for anything. So I cannot be incompetent because it has to be my management.”

    Incompetent? You are a jackass sir!

    Even IF I accept that notion that you have some sort of engineering background that gives you some expertise to comment along these lines. Your suggestion that I spend time exploring technology that has not even been extended in any major way outside of the large niche market has ZERO credibility.

    I have more important tasks to spend time on.

  12. oiaohm says:

    oldman the problem you have is you need to look up ánad its one of three words. “waste desert solitude”

    When you start doing multi words you need to be very careful on context. Waste and desert from ánad only apply to non living entities. A person or living in ánad is in solitude.

    Waste in Head is what you were going for. The correct line for the meaning you were intending one of the closest is “áhýðan beinnan héafod” there are a few others.

    Basically you have completely used the wrong word oldman yes wrong word makes it praise not a insult. Really do me a favour either get skilled at old english or don’t use it. Really there is a is an easy way to insult a person in old english. oldman sy unwís. Most are one word insults and I do have a stack of them.

    oldman maybe I am management the guy planning the future live cycles.

    “I am still bound by the requirements of my management and the application life cycles that exist here in my place of employ.”

    Those requires you can bet management will force on the new Arm laptops. That you will have to solve. Basically this is just an unwise excuse. Justification why you don’t have to be ready oldman.

    Major computer arch shifts don’t happen that often. Last one was over 20 years ago. You are still running as if there will not be a arch shift.

    The writing is on the wall informing you that a arch shift is coming. Of course some idiots will not be ready for it and get hurt.

    I love oldman my management is at fault not me. So I cannot be blamed for anything. So I cannot be incompetent because it has to be my management.

  13. oldman says:

    “Exactly how far away do you think the first Windows 8 arm laptops are oldman. Because that is how much time you have to have your desktop environment ready for arm. Then you will have arm in server room as well.”

    When ARM arrives is irrelevant to me. I am still bound by the requirements of my management and the application life cycles that exist here in my place of employ. This is something that you have a blind spot on. I suspect this cones from having an engineers orientation and bias. I would also suspect that you have either never been involved in long term IT projects and thus been exposed to dealing with the views and prejudices of those who are implementing on the platforms that you prove or you are simply omitting those experiences that contradict what you are trying to sell.

  14. oldman says:

    ““ánad beinnan héafod” Thanks for the praise. I do have solitude and clear thought in my head.”

    Actually that is one way to translate it. You may wish to consider the translation “Waste IN head”

    It is more accurate.

  15. oldman says:

    “As an educational organization you can likely threaten migration to GNU/Linux and get that other OS for $0 and fringe benefits if you wanted to maximize price/performance for your employer. In a couple of years every organization will have that option.”

    We are already well taken care of in that regard Pog, not only by microsoft by all our closed source proprietary vendors. Most educational institutions have had this for years.

    As I have already stated elsewhere. ARM Based systems are going to have to go quite along way before we have to even consider them. I expect that migration to be among the last projects that I tackle when I retire in 7-10 years.

    And I also expect microsoft to still be there as well.

  16. oiaohm says:

    Robert Pogson big one is arm you get more cores per area of silicon plus lower licensing cost.

    oldman the point is you have years now. If you leave it to the point of when institution has no option you are going to be in a world of hurt.

    Arm are not the RISC chips you know. http://www.heyrick.co.uk/assembler/riscvcisc.html

    Technically arm instruction set means its not risc pure. It has the best parts of cisc.

    Oldman it will turn true sooner than what you like. Possibly the first Windows 8 arm laptop or tablet you have that someone ask to run x86 internal applications you have on.

    Exactly how far away do you think the first Windows 8 arm laptops are oldman. Because that is how much time you have to have your desktop environment ready for arm. Then you will have arm in server room as well.

    Knowing up coming life cycles is critical to forwards planning oldman.

    Internally x86 is risc internal this was the only way x86 could keep up on performance. Some fancy code optimisation cpu level.

    The problem is time has moved on. Today we have profile guide optimisation of binaries and jit optimisation. In fact the x86 cpu guess me profiling works against 100 percent effective profile guide optimisation or jit optimisation.

    So yes the very thing that allowed x86 to beat risc the cpu optimisation of code is the very thing pulling x86 straight to the bottom.

    Tech has moved on x86 has lost the battle against risc. Attempt to beat risc is why x86 dia size is blowing out and why profile guide optimisation does not work well on x86. Send x86 cpu an optimised binary incompatible with what the x86 cpu optimiser expects and the processing can slow down by the factor of 20. Risc processors don’t have this issue where the cpu optimiser and produce binary code can fight.

    Arm has focused on binary size. Reason why thumb2 instruction set exists. 16 and 32 bit instruction set that perfectly co-exists. There is also a 16 32 and 64 bit instruction set in arm that perfectively co-exists. No context switching to change modes.

    Now what is wrong here. Risc cpu design says your instructions should be all the same length. Arm is not Risc in this regard but closer to CISC.

    Yes the space saving CISC got by mixing different length instructions arm has as well.

    X86 is at the end of what it can technically provide us with. Runtime optimisation by cpu has turned out not to be such a hot magical idea.

    “ánad beinnan héafod” Thanks for the praise. I do have solitude and clear thought in my head.

  17. oldman wrote, “planning the return to the RISC based systems that I left.”

    You did not leave RISC because RISC was less inefficient. You left RISC because Moore’s Law allowed Wintel to ship more units at lower prices. Now ARM under FLOSS allows the world to ship more units at lower prices. ARM is taking just a few cents per CPU for billions of CPUS. The chip fabs are taking just a few dollars per CPU. FLOSS is taking $0. The OEMs are making a reasonable margin because there is competition in the segment, something Wintel killed decades ago by illegal means. Fortunately for you Wintel is starting to compete on price/performance and may yet remain feasible in IT but no longer be a monopoly. As an educational organization you can likely threaten migration to GNU/Linux and get that other OS for $0 and fringe benefits if you wanted to maximize price/performance for your employer. In a couple of years every organization will have that option.

  18. oldman says:

    “This is the problem oldman the writing is on the wall. You can kick and scream all you want. In the end I will be proven right and you will be stuck with the problems. Same idiots existed when the mainframe age ended. It cannot end desktops cannot take off. Yep a few years latter what happened.”

    I could point out that there is quite a big difference between then and now. I could also point out that the mainframe is still around and thriving.
    But I who actually works in enterprise IT am the idiot, while the ánad beinnan héafod nym who does itinerant IT in rural Australia is not.

    Spare me!

    Who is proven right is simply not relevant sir. What is relevant are the policies, deliverables, and project life-cycles of the institutions that will actually use and implement your new winder technology. The fact remains the even IF your speculative time line turns out to be true, there is a good chance that it will be quite a few number of years after that before my institution feels the need to look into converting to it.

    When and ONLY when that day comes will be when I actually have to deal with assisting in planning the return to the RISC based systems that I left.

  19. oiaohm says:

    oldman as you admit wine is not good. So ending up in a location where you are forced to use it is foolish right. This is why I see you as foolish oldman.

    Difference between you and me. I see what AMD and Intel and the big hardware makers are up to.

    AMD is moving to support ARM Chipsets more. Nvidia is already supporting Arm chipsets. We have new players on the block like ZiiLabs ZMS-40. Intel video card is already found inside arm.

    The writing is on the wall for us who can find where the wall of what is coming is. Not just twits like oldman who are only look at what is here.

    Most new devices in production today were planned at 2 to 3 years in advance. So yes there is a 3 year in advance crystal ball. What I can clearly tell you is no where in the next 3 years is there a super powerful x86 chip. Dia size is bring x86 undone. There are some super powerful arm chips there are some super powerful mips chips. IBM makes a little showing in power due to enhanced cell engine.

    Only x86 core part maker not to announce some reference to working with arm is in fact intel.

    This is the problem oldman the writing is on the wall. You can kick and scream all you want. In the end I will be proven right and you will be stuck with the problems. Same idiots existed when the mainframe age ended. It cannot end desktops cannot take off. Yep a few years latter what happened.

    No change in the IT world comes from no where.

  20. oiaohm says:

    Phenom those changes you talk about are not that hidden. http://test.winehq.org/data/ Wine does not depend on MSDN documentation.

    Wine depends on a method called BlackBox reverse. This is one huge test suite. That finds defects in MS ABI implementation that explains why particular old applications don’t run on new versions of windows. Currently only 2008 does not fail Blackbox checking somewhere.

  21. oldman says:

    “@ldman doesn’t think, he just repeats himself. I guess you’re going to take my advice and try to bore me to death. Good choice. Judging by your complete lack of creativity that’s your best shot.

    “FOSS rules because FOSS rules…”

    “You need to get some respect for FOSS…”

    You seem to suffer from the same problem. Actually your real problem is your zealotry and blindness. Coupled with the fact that you butt in to conversations that are not addressed to you personally. You create your own boredome because you refuls to accept delivery on the notion that your beloved FOSS might not suit all cases. You refuse to accept delivery on the reality of fosses limitations even in the mundane tasks that you outline. You proclaim that I cant work without FOSS, yet when called on it you say nothing.

    At least Pog defends his position. All you do is Jeer at the “ebil trolls”

    IMHO, this is not only boring, but stupid!

  22. Phenom says:

    Pogs, you deliberately chose not to understand. Shame on you. Let me humbly try to bring you to the right track.

    Wine tries to reproduce the API of NT with some reverse engineering and based on the front-end documentation of MSDN. MSDN describes how to use the API, not how it is implemented. However, the Wine team always has to return to API they thought they covered, just because of fixes in future releases. MS are making a hell of a lot of tweaks to ensure compatilibity with existing software systems, and that involves hidden changes in the implementation of the API. Mind you, these are transparent for developers, documentation does not change, and the binaries just keep working. However, Wine are never aware of these, unless they go over and over again the code to debug and realize what is going on behind the scenes.

    So much for your troll attempt, try again.

  23. oldman wrote, “Winw will never and can never be a complete emulation of the windows API’s and because microsoft is under no obligation to support the efforts of wine to emilate their API’s wine will always be a day late and a dollar short as far as windows compatability.”

    So much for the stable API of that other OS we always hear about in this blog…

  24. Kozmcrae says:

    “Run along sonny…”

    “Bu$ine$$ is Bu$ine$$”

    “Applications, applications”

    @ldman doesn’t think, he just repeats himself. I guess you’re going to take my advice and try to bore me to death. Good choice. Judging by your complete lack of creativity that’s your best shot.

  25. oldman says:

    “You may be an adult and you may be talking, but you lack the wisdom of an adult and your words lack integrity.”

    Yeh, yeh Whatever…

    Run along sonny…

  26. Kozmcrae says:

    “Run along and play sonny, the adults are talking…”

    “If it doesnt run what I want it to run, its basically worthless.”

    @ldman is speaking for the whole World again. If it’s worthless to him, then it’s worthless to everyone else.

    You may be an adult and you may be talking, but you lack the wisdom of an adult and your words lack integrity.

    You’ll have to try a lot harder than “run along sonny”. You’ll have better luck if you try to bore me to death. You’re much better at that.

  27. oldman says:

    “I guess there are idiots at IBM”

    A fact that I can actually attest to having dealt with IBM directly for over 30 years….;-).

    The fact that IBM has recognized the existence of wine (or “whine” as i call it in my less charitable moments after dealing with it) says nothing about its ultimate quality. Lets face it Pog, wine is a hack propagated by those like yourself who no longer want to run windows but who have windows based applications that they can’t do without. Winw will never and can never be a complete emulation of the windows API’s and because microsoft is under no obligation to support the efforts of wine to emilate their API’s wine will always be a day late and a dollar short as far as windows compatability.

    The fact that some commercial company can sell me third party support for a commercial application is meaningless to me. They are not the creator of the software they are running and they can not guarantee me that they will be able to fix a glitch in a timely manner of an update or patch to the software in question “Breaks” its ability to be supported under wine.

    It is far easier for me to run Linux in a virtual machine under windows than it is to run my applications under IMHO some piece of crap like wine.

  28. oldman wrote, “only an idiot would recommend a hack like wine in an enterprise context. And you expect me to take your advice seriously?”

    I guess there are idiots at IBM:
    “There might be applications left that do not lend themselves to be moved off of a Windows client using any of the methods that we have mentioned. If these applications are to be used after the client migration because of economical, legal, legacy, or other reasons, a solution has to be found.

    At present, it is possible to use a Windows emulator (such as Wine) to support running Windows applications on a Linux-based client. “

    See Linux Client Migration Cookbook, Version 2
    A Practical Planning and Implementation Guide
    for Migrating to Desktop Linux
    page 102

    Thousands of applications are on Wine’s “Platinum” list, including PhotoShop 7.

    IBM has a huge business but one part of it is helping businesses migrate to GNU/Linux. They know how to do it. Wine is a last resort but it works. I have used Wine a few times when some teacher brings in some application and asks me to make it run on GNU/Linux. That has only happened two or three times in a decade, however. Mostly they are overwhelmed by the applications available in the menus.

  29. oldman says:

    “oldman the simple fact is most likely not all the vendors you use will convert across to arm.”

    Really? If you can tell me that I would like you to guess lottery numbers for me!

    “You really do have no option bar to start now. Locating all the legacy long since deceased vendor software. That is no longer going to work.”

    Again which deceased vendors are you talking about? What is no longer going to work?

    “2 to 3 years is normally enough time to weed the crap out the system oldman.”

    What weeding do you think I can do sir?
    I do systems design in support of ERP applications that have been chosen by my institution based on their utility for a particular task. As has already been noted, the lifecycle of these systems can and will span multiple generations of hardware.

    “The thing is you want to start conversions while you still have options and time oldman. This way when they come they are fairly painless.”

    In point of fact I am not going to start anything based on the biased speculation of someone sir. I have other futures projects on my plate of more immediate importance to my management.

    “You know x86 programs will not run on Windows 8 arm. This is a known given.””

    But I will bet that if ARM becomes the hit that you think it will become outside of mobile devices that those vendors will offer ARM platforms certified to run on windows 8. It is also a good bet that those same vendors will make it their business to ease any transitions for their customers.

    ” If the vendor is dead or gone on a item you are using. It is stuffed by today’s standards.”

    perhaps you with you penchant for picking no name commercialized FOSS companies have seen this, but I have yet to have to deal with the abrupt orphaning of a product in use.

    “You cannot think that it will work on the future Windows platforms.”

    Why Not. Because your bias says it won’t?

    “oldman software that will not run is worthless long term. I stated the real value of it.”

    No you stated your bias. and I responded politely that I didnt think much of your speculation.

    “Software that you currently own is worth nothing. Only thing that makes it worth anything is the future versions that will keep you working as stuff changes.

    Thats funny, I periodically run my 22+ year old copy of scorePC to generate a copy of the score that I created using it. The company is long gone yet it still runs. That still gives it value in my eyes.

    “What oldman has told me clearly is that the Vendors think for him. “vendors whose software we use” This is just his current vendors.”

    If this is what you hear, then you have a comprehension problem. Our vendors provide solutions with a given set of function and feature that meets the needs of the our institutional stakeholders (i.e. the Users). We get technical input into the selection process, and we can and have had to recommended against a particular product. But absent a real show stopping technical issue, once a package is chosen, we get to implement it and support it as long as it is needed.

    And you Mr. Microsoft VAR get to do the same ON WINDOWS, when you are told to. by YOUR management as well.

    “Is there not the possibility that due to the arm change that you drop a few vendors and replace them with new vendors that don’t care about OS. This is thinking for your own companies interest not the interest of vendors.”

    Unless the follow on ARM based product is so superior in function and feature that the users ask for it’s implementation, there is very litle possibility of that sir.

    “This is why it takes 2 to 3 years in a good run. Not only to change software but to evaluate if any of your current vendors should be given the boot and replaced. Of course there is a risk that some software is too in grained to be replaced this is where items like wine will come in.”

    wine?

    ROFLMAO

    To put it bluntly, only an idiot would recommend a hack like wine in an enterprise context. And you expect me to take your advice seriously?

    “Oldman basically wants to leave it to the last min. Head in sand method harms companies oldman.”

    No I deal with reality. The possibility of having to support legacy systems long past there end of life is one of the reality of enterprise IT. The fact that you can characterize my conservative philosophy, which is in turn driven by requirements not of my own making, tells me that you have never actually had to deal with the realities of Enterprise IT, except possibly as one of the unwashed geeks in the back room.

  30. Phenom says:

    “Run along and play sonny, the adults are talking…”

    Assuming you can call our beloved Ohio an adult. 🙂

    “This is why it takes 2 to 3 years in a good run. Not only to change software but to evaluate if any of your current vendors should be given the boot and replaced”
    Ohio, you and Pogs might be quite surprised here, but software at an enterprise is something rather conservative. Many enterprises run systems as old as 20 years, and have little intention to replace them. Because they work.
    I was in a Novotel hotel in Wolverhampton a few days ago and incidentally I found out that their hotel software is running on… DOS. A text-mode DOS network application.

    And, it works. The interface is quick to use with appropriate shortcuts, and they can process a check-in and check-out in less than 2 minutes each, including issuing room cards and invoices.

    Try tell them replace their software.

  31. oldman says:

    “Worthless to whom?”

    Run along and play sonny, the adults are talking…

  32. oiaohm says:

    oldman the simple fact is most likely not all the vendors you use will convert across to arm.

    You really do have no option bar to start now. Locating all the legacy long since deceased vendor software. That is no longer going to work.

    2 to 3 years is normally enough time to weed the crap out the system oldman.

    The thing is you want to start conversions while you still have options and time oldman. This way when they come they are fairly painless.

    You know x86 programs will not run on Windows 8 arm. This is a known given. If the vendor is dead or gone on a item you are using. It is stuffed by today’s standards.

    You cannot think that it will work on the future Windows platforms.

    oldman software that will not run is worthless long term. I stated the real value of it.

    Software that you currently own is worth nothing. Only thing that makes it worth anything is the future versions that will keep you working as stuff changes.

    Oldman you are assigning value wrong.

    What oldman has told me clearly is that the Vendors think for him. “vendors whose software we use” This is just his current vendors.

    Is there not the possibility that due to the arm change that you drop a few vendors and replace them with new vendors that don’t care about OS. This is thinking for your own companies interest not the interest of vendors.

    This is why it takes 2 to 3 years in a good run. Not only to change software but to evaluate if any of your current vendors should be given the boot and replaced. Of course there is a risk that some software is too in grained to be replaced this is where items like wine will come in.

    Oldman basically wants to leave it to the last min. Head in sand method harms companies oldman.

  33. Kozmcrae says:

    “If it doesnt run what I want it to run, its basically worthless.”

    Worthless to whom?

  34. oldman says:

    “You really have no option bar to start work now.”

    HAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA….

    You really do take yourself way too seriously sir.
    In point of fact my institution has plenty of options before we have to look into such a wholesale conversion, and regardless of its success any such conversion will be we driven by the availablilty of ARM support for vendors whose software we use.

    As far as your estimate of the value of my personal productivity software is concerned, I think you know what I think of that, Mr. Microsoft VAR.

  35. oiaohm says:

    oldman this is the problem. Person like me will be able to use these super powerful arm chips without issues. Same with a person like Robert Pogson.

    You on the other hand will have to pray that MS stuff is provided.

    Arm is also making there own gpu mali. That is also Linux friendly. Nice bit about mali is that it renders straight to a memory buffer so you can simply convert opengl output to vnc/splice/what ever method to get output to thinclients so yes you do have functional 3d on thinclients quite light.

    Question is the chip worthless or is the software you using going to be worthless. oldman.

    Oldman you are stuck between a rock and hard place. X86 is not going to be the power house cpu any more. The power house will be arm. So you might end up mixed. With like your web servers, databases …. on arm. Particular desktops still on x86.

    Really you need to have a real close look at what you have in the server room that is x86 locked. That is going to be trouble. Arm licensing is cheaper than x86 licensing. So yes you are going to have price problems as well. Price to Power is going to be in the arm favour.

    The wintel time is over. All that is left is the screaming from people like you oldman who think that closed source software is a good investment. As you become stuck with tech that cannot compete and just keeps on getting more expensive.

    Result of sticking on x86 will be higher power bills and lower performance.

    A 512 chip is that far over powers that most of your existing applications will run quite well in qemu with fpga assistance. But you will want to be running as many native applications as you can.

    So yes oldman if I was you I would kinda be making sure that feature requests exist in libreoffice and everywhere out for what you really require.

    The end of road is coming up fast.

    Please note Mono .net does not run on ARM. So all that asp.net stuff. That businesses have laying around is mostly worthless.

    The wintel run had to end at some point.

    It takes about 2 to 3 years to migrate platform.

    You really have no option bar to start work now.

    Windows to Arm will break all your x86 applications anyhow. Windows to Linux on arm you do have the option of using qemu with wine to run some of those older windows x86 applications.

    If you start you run too late oldman you may have no other option bar the Linux path so your key applications work. I worked out this problem coming a few years back oldman. So you better be testing your key applications with wine so you have fail back location.

    Arm public statement about wanting desktop and server is what tipped me off. I know arm they do anything they set mind to.

    Yes I think you have the statement work. Your software is worthless oldman you just have not woken up to that fact yet.

  36. oldman says:

    “Now 2 of the 2014 chips in desktop machine would kinda rock. One for video card and one for main processing unit.”

    If it doesnt run what I want it to run, its basically worthless.

  37. oiaohm says:

    oldman beware of another thing when arm moves down to 14 nm that if in time table will be by 2014. This machine moves from 128 cores per chip to 512 cores per chip.

    By end 2014 most workloads without redundancy will be able to be replaced by a 1U unit running arm. With most likely a rack of disks. fpga unit can be programmed for helping to convert x86 code.

    Of course you will want redundancy. So 2x1U + drive racks. If you are needing bigger than that you are google or facebook or a render farm.

    The shocking part is the density of cpu’s in these arm units is approaching gpu density of cores. Basically with chips like this why go opencl its most likely slower than adding more arm cores. GPU cannot allocate ram this brings overhead.

    You can fit about 10 cpu chips correctly placed inside a 1U case. 5120 cores per 1U in 2014. When you blade that number goes up massively. 2U case you are looking at 20 000+ in 2014. Most likely cooling and power feed is going to become a limiter. Really I look forward to running blender on a 20 000+ puppy.

    Now 2 of the 2014 chips in desktop machine would kinda rock. One for video card and one for main processing unit.

  38. oiaohm says:

    oldman Microsoft is working on arm version of windows server. Particularly with the big HP arm based servers coming out.

    The current fight over EFI is to allow windows to boot on arm hardware. Linux has its solution using uboot and others to pass the Linux kernel a arm hardware description so able to use a generic arm kernel with a initfs filesystem containing the drivers for basically everything.

    Arm hardware is not like x86. Its not probable. So unless the boot loader tells you what is there you are stuffed. Arm chips by default design don’t have serial numbers or any other ID marking you can request to find out what they are. cpuid that you have under x86 to tell you want the processor is don’t exist under arm. None of your items on the cpu busses contain any method to id what is there in advance either.

    Also just to be really nasty some people would think lets just thump functions to find out what I have. Between arm versions they moved the catch method for invalid functions yes and try to hook to invalid function hook that does not exist is a invalid function and locked the cpu up if you don’t have a invalid function hook already connected. Completely forget every method of probing you know. All probing methods are land mined on arm.

    There is 64 bit arm. Holds it own quite well particular-ally with the fact they normally have a section of fpga you can program custom complex maths functions into so they can be done in one clock cycle.

    Yes the lack of ID’s on arm is going to cause MS genuine advantage system some major headaches.

    http://www.anandtech.com/show/5027/appliedmicro-announces-xgene-arm-based-socs-for-cloud-computing

    Basically this is a Arm tileish design chip. Yes its fully 64 bit. 3ghz 128 cores in a single chip.

    Really 128 cores inside a tablet I think not. 256W power draw in a tablet I think not somehow.

    Yes those puppies do interconnect properly where the tiles don’t fully. So yes you can do a numa configuration of the x-gene.

    It will not take long for power density in a 1U to increase massively. 200+ cores to a 1U will become normal. This kind of chip could easily fit 1000+ per 1U box on a flat motherboard. 2U you can take these suckers blade.

  39. oldman says:

    “The power you are talking about oldman is less than what is in 1 U tile system. With Microsoft coming to Arm as well at some point you will be forced to give up the x86 crap.”

    Tilera is an interesting technology. it has its place and I suspect it will do well. However I think that its effect if any is far enough in the future as to be a non issue for me and I suspect a large number of people as well.

    “The age of x86 is ending. The age of Arm is starting.”

    I think that x86-64 us too useful and to versatile to be going anywhere soon. ARM in comparison is a jumped up embedded controller that shines in low power tasks on portable devices, but its ability to swcale out of that niche is questionable at best. The promise that ARM has designs with higher processor counts means as little to me as windows 8 means to you.

    The significance Microsoft coming to ARM at this point is questionable. They clearly want a piece of the tablet market, but the utility of the ARM based windows 8 implementation beyond that is questionable,

  40. oiaohm says:

    Arm chips exist in blade combination. Blade is the scare 3900 to a 2 U 19 inch box. Same volume of arm blades will basically eat x86 blade for lunch for what they can process.

    x86 – GPU blades combinations sorry to say still a bit weak. You are using a huge volume for the processing power you are getting.

    Even that the arm’s don’t have very big GPU unit on them. They still have a small opengl 2.0 SE GPU every 4 cores. Software exploiting arm 3900. 975 small GPU units in that 2 U 19 inch box. Even that they are not the fastest volume kinda wins.

    Tilera is a different beast. Its normally the module you plug into the switch as switch base load balancer.

    http://www.tilera.com/sites/default/files/images/content/tile_empower_gx_diag_large.png

    This is where it gets evil. See the optional PCI-e card.

    Inside the tilera platform you go as bad as it sounds 200 CPU to a 1 U rack. Yes that is 8 10 Gb per sec connections network connections. That is kinda great for network disc IO for some reason. Yes they can maintain 100 load on 8 10 GB ports no problems.

    Compare the to high density arm tilera might appear weak. The tilera better designed cpu interconnects do come into there own when handling web traffic.

    The power you are talking about oldman is less than what is in 1 U tile system. With Microsoft coming to Arm as well at some point you will be forced to give up the x86 crap.

    The age of x86 is ending. The age of Arm is starting.

  41. oldman wrote, “it is far easier to gang together piles of x86 blades or GPU blades.”

    That may currently be true but Tilera and ARM are headed to much higher computing density with lower power consumption. These folks are going to provide tools to make the change in architecture rather transparent. The kernel drives the hardware and the rest is the same as any GNU/Linux system. Don’t forget, the x86 solution requires fans. They take up a lot of space. The new systems will still need fans but each fan will cool a lot more CPUs.

  42. oldman says:

    “It is the best at what it does but since windows will not run on it you end up with it as a expansion card.”

    I suspect that tilera will spend its days in a niche market as the engine for high performance hardware appliances. HPC will give it a pass because it is far easier to gang together piles of x86 blades or GPU blades.

    At the same time an intel theoretical limit of 50 cores per CPU doesnt really bother me. I am installing 4 way 10 core as VMWare visualization engines based on IBM 3950 X5 boxes. I can scale to 8 Way via numa if I need to. It will probably be quite a while before we come to need that density.

  43. oiaohm says:

    oldman 100 core x86 has been attempted by intel.

    50 was max limit intel was able to get to. Past that the silicon is getting too big to be stable and running into major heat problems.

    The change over to arm become more dominate in the server room in the next few years is because x86 is basically at its max scale.

    Its the translator from x86 CISC to internal RISC that kills you. That is in fact in a lot of x86 chips takes up more die space than the RISC core itself sometimes by factor of 2. 50 core proto intel did as a risc would have been 100 cores +.

    oldman no
    “assuming that they did using the clean room reverse engineering techniques, that would be another thing entirely.”
    x86 is patent locked. Unless you are licensed to the patent pool you cannot implement in hardware.

    Best hope for tile would have been like a transmedia x86 to tile conversions program. Then you would have still had the issue of windows not really like 100 cores in a single die.

    When you know the issue we must get of x86. The licensing of x86 is blocking development.

    Tile is in business because it is good. For particular workloads Tile is used as a coprocessor in a windows machine. Basically hack. It is the best at what it does but since windows will not run on it you end up with it as a expansion card.

  44. oldman says:

    “Basically we need a FOSS OS core as dominate so more hardware types can compete with each other.”

    Perhaps, but the reality of commerce is that the profitable good enough drives out the best.

    Besides, a 100 Core RISC chip is useless to the “average” user. Had Tilera been able to bring out a 100 core x86 clone chip, assuming that they did using the clean room reverse engineering techniques, that would be another thing entirely.

  45. oiaohm says:

    Tilera is the example of what is wrong with the Microsoft model. There make a great powerful cpu. Yet they can forget having Microsoft support because there chip is in too limited production.

    Basically we need a FOSS OS core as dominate so more hardware types can compete with each other.

  46. M$ is pretty well irrelevant in this. I doubt they will release “8” for Tilera. Tilera does not meet their “certification requirements”/limitations. Debian GNU/Linux will likely be suitable, but of course there may need to be special drivers for the chipset. I suspect Tilera has that stuff ready. Debian GNU/Linux is the most popular web-serving distro, I’ve read.

    at the beginning of 2012, Debian was used by 29.4% of all Linux-based sites (and by 9.7% of all web sites), while CentOS was used by 29.1% of all Linux-based sites (and by 9.5% of all web sites). Debian is also the fastest growing operating system at the moment: every day 54 of the top 1 million sites switch to Debian,

    There are good reasons for Debian’s popularity. APT is number one.

  47. Conzo says:

    You forgot to bash Microsoft and to mention Debian GNU/Linux 🙂

Leave a Reply