1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Hardware AMD Betting Everything on OpenCL

Discussion in 'Article Discussion' started by arcticstoat, 30 May 2011.

  1. DbD

    DbD Member

    Joined:
    13 Dec 2007
    Posts:
    494
    Likes Received:
    10
    From what I can tell they are saying on one had they need opencl to take off to get the best of their cpu's but on the other they aren't actually going to do much to make it happen - that's down to the mythical "developers".

    The reason that CUDA took off is because nvidia spent a lot of £££ actively developing it - CUDA is several years ahead of opencl both as a standard (i.e. basic instructions available), in the sdk (i.e. the secondary libraries needed to do interesting stuff), in the development environment (i.e. debuggers, etc) and in training/support.

    That needs to happen for opencl too - if AMD just leave the "developers" too it then it'll take years and years. No one else is really pushing it - nvidia have cuda, Intel still want everyone to use x86, the rest of the world is more interested in tablets and ARM (e.g. MS). It simply won't happen. Even if AMD did push really hard it'll still take several years - far too late for llano or perhaps even it's successor.

    It is quite telling that in the interview he basically said they are abandoning high end computing to CUDA. Currently that is the only market today where gpu compute is essential (all new super computers seem to now come with it) - it's also one of the highest profit markets too. It is also the source of much future innovation in normal PC's - what happens in super computers filters down to the rest of us. Why are they abandoning that? I can only assume because they really don't have a proper plan to attack the gpu compute market - just a few marketing guys and a lot of hope that it'll magically take off without any real work by AMD.
     
  2. Kaihekoa

    Kaihekoa New Member

    Joined:
    26 May 2010
    Posts:
    24
    Likes Received:
    0
    These guys are both blowing a bunch of marketing/PR smoke. They can't compete with Intel performance-wise. In the last few generations of processors all they've done is crank up the clock speed and sell units for a net loss. So they are toting this OpenCL as what's going to make them competitive again, but I doubt it will happen anytime soon. They cite examples of web browsers and flash using GPU acceleration, but that's not OpenCL. Most IE users are still using 6.0. Why? Because on average most computer users aren't knowledgeable about tech hardware or software. Now these same people are supposed to be the driving force of AMD's new OpenCL and GPU acceleration?

    They need to get their R&D in a think tank and come up with a CPU architectures that will put them ahead of Intel without having to mask their inefficiencies with PR. This article having to go all the way back to the Pentium 4 for a time when this company was competitive is indicative of their present ability to succeed in the market. Nvidia's GPU programming language, CUDA, isn't going to disappear. Nvidia invested a lot of money to get development started, and now corporations and academia use it for supercomputer level projects. Nvidia's Tesla cards are better equipped for those tasks so they won't be throwing away all the money they invested for OpenCL anytime soon.

    And before you go screaming fanboy, my desktop uses a Phenom II X4 and Radeon because they offered good performance at a low price, but even an AMD user has to acknowledge that their processors have been in Intel's wake since the Core architecture was developed. Ideally, it would be better for consumers if the disparity in performance per clock was evened with AMD's Bulldozer as it would keep Intel's prices down.
     
  3. knownballer

    knownballer New Member

    Joined:
    8 Mar 2009
    Posts:
    13
    Likes Received:
    0
    Amd certainly is betting a lot on the standard, and though the PR guys didn't really say it, they really don't have much of a choice. Intel and Nividia both took a big advantage years ago by purchasing Havok and Ageis respectively (funny we don't hear much about Havok anymore...) and as the discussion moved towards using the gpu architecture as a processing platform, this really was the only option available.

    Indeed AMD needs to do this, but the simply don't have the resources to. Intel a few years ago gave my school 300 million for research on multi-cpu computing. Likewise Nvidia has much more room to investing in building CUDA while AMD struggled to get back in black.




    I don't think you've been paying close attention:
    http://www.engadget.com/2011/05/28/amd-ships-five-million-fusion-chips-says-its-sold-out/

    Amd has been the main driver of the whole fusion ideology while Intel has been more concerned with pushing x86 into everything. I would also say that Amd has been the better chip architecture designer for the past decade but intel has been able to leverage their fabrication advantage, as well as a few dirty tactics to gain an advantage over Amd.

    But your assertion of the consumer being key is precisely why though OpenCL is key for Amd's long term plans, they may not need the platform to be adopted to sell APUs. We're reaching a point where for consumers the cpus are "good enough" and the real key is gpus power. That's why bobcat has been so successful. And as time moves the OpenCL will be adopted. But the big problem is if that is the future, there is another company in Arm that is is a great position to expand in the mobile space.

    I think the key a lot of people have been missing is that Amd's aim isn't just to add a GPU onto the Cpu chip like a SoC, but to integrate elements of the architecture onto the Cpu like many elements have been added to the Cpu over the years. But for this to work the Gpu part cant communicate the way is does now. I don't know the details but the Gpu just isn't seen by a computer in the same way a Cpu is and that is affecting how software is able to take advantage of their abilities. Just like multi-core computing, the key will be on the software side and how programmers will take advantage of the architecture. There's so many factors there's really no way of tell how things will turn out a few years from now. Who knows, maybe Arm might move to become THE architecture moving forward but right now I think Amd looks like the have the best vision moving forward.
     
  4. PQuiff

    PQuiff New Member

    Joined:
    7 Sep 2001
    Posts:
    557
    Likes Received:
    4
    Im not particulary bothered about OpenCL.

    What i want from AMD is Hardware. I dont care if it will accelerate my IE, or excel. When i fire up battlefield 3 i want to know that my AMD CPU and GPU is faster than the equivalent Intel/Nvidia setup.

    AMD. If your bulldozer CPU is slower than the intel one, i wont buy it. If your GPU is slower than the Nvidia one i wont buy it. I worried that AMD is playing catch up, rather than saying "let create hardware that will nuke our competitors" there planing for hardware to match there competitors. And by the time they release it its old hat.

    Id say its all or nothing on OpenCL, but its all or nothing on your next hardware release to.
     
  5. fingerbob69

    fingerbob69 Member

    Joined:
    5 Jul 2009
    Posts:
    801
    Likes Received:
    16
    Ask yourselfs who amougst the big movers and shakers use AMD/Ati gear in their devices and who could, therefore be a main driver of this?

    Answer ....Apple.
     
  6. blackerthanblack

    blackerthanblack Active Member

    Joined:
    17 Sep 2004
    Posts:
    670
    Likes Received:
    27
    Not so. Most computers are on IE6 because of business. This is the lowest supported version by Microsoft - there's no point causing potential headaches upgrading hundreds of company PC's if it's not necessary.

    This will soon be changing as I believe IE6 support is being removed. Expect most PC's to start using a newer version soon.
     
  7. Snips

    Snips I can do dat, giz a job

    Joined:
    14 Sep 2010
    Posts:
    1,940
    Likes Received:
    66
    I ain't the one whining and moaning. AMD are saying doing it their way is better because.....it's better on their playform in readiness for their furthcoming product releases. They aren't saying this to promote the openess of OpenGL, they are promoting it on their products.
     
  8. maverik-sg1

    maverik-sg1 Member

    Joined:
    18 Aug 2010
    Posts:
    363
    Likes Received:
    1
    I think it's right to promote an open standard and OPENCL will be compatable by iOS etc.. not just windows PC's, what abouts arms intention to grab some desktop action? How will that effect the OPENCL war?

    Also consider for the PC, DX11's DXcompute is an open standard so, whoever is king of the DXC castle, is it safe to assume they will also own (based on current GPU/APU products) the OPENCL castle too?

    I so want AMD to succeed with all the tech they will release this year and beyond - but I think there are bigger fights than the one for OPENCL supremacy, which wil of course be supported by all the market players.

    It's important, but probably only accounts for 5% of the battles they need to win right now...... after all, GPGPU computing and (more importantly) mainstream programs that use it are in their infancy right now - Bulldozer and Llano will probably be replaced by the time OPENCL performance will effect our purchases.
     
  9. iatacs19

    iatacs19 New Member

    Joined:
    31 May 2011
    Posts:
    2
    Likes Received:
    0
  10. TheUn4seen

    TheUn4seen New Member

    Joined:
    22 Oct 2009
    Posts:
    68
    Likes Received:
    2
    It's a nice interview, but the most obvious conclusion is that Bulldozer will be a disappointment.
    I can see a lot of problems before openCL will become a mature, versatile standard ready to be adopted for production enviroments, and what I care about now is that I need a new CPU. And I need it to be fast. I don't care about GPGPU since right now it's limited to several applications (and none of the ones I use works with openCL - only CUDA if at all).

    Also, gamers don't care about GPGPU, they need a fast GPU paired with a fast CPU, not super-mega-killer GPU paired with crappy CPU that needs openCL to get decent performance - since even now most games are single-threaded (and the ones that aren't mostly use two threads, in an era of cheap six and eight core CPUs) I don't see dual-GPU (openCL "APU" + dedicated rendering GPU) games inany foreseeable future.

    I know it's not a place to talk about Bulldozer, but all this openCL and GPGPU talk and wishful thinking seems to say it all - AMD can't deliver on CPU front. Which is a damn shame since I had high hopes for Bulldozer.
     
  11. maverik-sg1

    maverik-sg1 Member

    Joined:
    18 Aug 2010
    Posts:
    363
    Likes Received:
    1
    We can't write it (Bulldozer) off until it's been released.

    Like multi-threading in games, off loading some parallel instruction sets in games may also boost performance - the challenge of course is getting developers to do that, which essentially means the cost has to have a benefit to them (probably a cash injection) and overall it will have to be a 'selectable' feature to satsify the whole market (those without GPGPU/APU) - although, if/when consoles engage in this style of development then it will be a win/win for everyone, purely becuase after a few years consoles need every bit of perfromance squeezed out of it's architecture and it wil be just part of the multi format (xbox720/PS4/PC etc...) programming.
     
  12. mischasan

    mischasan New Member

    Joined:
    1 Jun 2011
    Posts:
    1
    Likes Received:
    0
    Hahaha! OpenCL is an alternative to CUDA? OpenCL is a *wrapper* for CUDA. It's just a glue layer that Khronos worked out, broad enough that you could conceivably use the same code on AMD/ATI or Nvidia hardware. All the working language is straight CUDA code. Fine. I use OpenCL because that's what you get on MacOS X; I bought a Mac because it has two GPU's and *the good one is an Nvidia GPU*. I can run my OpenCL code where there's no GPU available, to check that it compiles and functions. I can run it with an AMD GPU, But Nvidia and AMD hardware are worlds apart; you don't even bother with the same algorithms. So what if AMD chips increase their FLOPS rates? I can use shared-memory in an Nvidia GPU to do serious code.
     
  13. xinaes

    xinaes New Member

    Joined:
    17 Jan 2011
    Posts:
    103
    Likes Received:
    2
    OpenCL kernels aught to be able to run on all kinds of processors, as long as the right drivers are present. In some cases OpenCL running on a standard CPU performs better than old fashioned CPU code, from what I've read. Workloads can be arranged at runtime such that programs adapt to whatever processors are available, distributing the work across all cores seamlessly (at least in theory).

    It would be good to see more high-level programming tools and libraries for OpenCL, though... nVidia may support OpenCL to an extent, but the capabilities of their Parallel NSight tools and such like are not as strong as they are for CUDA. I suspect they are deliberately holding off for now so they can continue to encourage people to use CUDA, but will move quite quickly if (hopefully when) OpenCL becomes more established and rival tools are in evidence.
     
  14. knownballer

    knownballer New Member

    Joined:
    8 Mar 2009
    Posts:
    13
    Likes Received:
    0
    Amd is a business

    It all depends on what is considered a failure. Keep in mind after Amd spun off it's fabs, if changed it's focus on not competing at the highest end, but the point when the most cpus are sold. You may not like this, but it's sound for a business standpoint, in fact it worked for Ati and now their gpus are eating nvidia's lunch. Gamers only count for a small part of the market. And you should care about ggpus. They create a bare minimum standard for video processing that will no doubt trickle down to developers.
     
Tags: Add Tags

Share This Page