OpenCL should be able to run on x86 ... but probably not fast

Article: Intel's Sandy Bridge Graphics Architecture
By: David Kanter (dkanter.delete@this.realworldtech.com), August 16, 2011 7:49 am
Room: Moderated Discussions
Mark Roulo (nothanks@xxx.com) on 8/16/11 wrote:
---------------------------
>David Kanter (dkanter@realworldtech.com) on 8/15/11 wrote:
>---------------------------
>>Mark Roulo (nothanks@xxx.com) on 8/15/11 wrote:
>>---------------------------
>>>David Kanter (dkanter@realworldtech.com) on 8/10/11 wrote:
>>>---------------------------
>>>>AFAIK, Intel has not exposed any shared memory to SW, which is required for OpenCL.
>>>>They could use the L3 cache for shared memory, but the performance seems like it
>>>>would be pretty awful due to high latency.
>>>
>>>Shared memory is a logical, not physical, concept in a cache-coherent system.
>>>The L1 would probably wind up being used for typical OpenCL codes (the mapping being one nVidia SM -> 1 x86 core).
>>
>>I was speaking of the Sandy Bridge GPU.
>
>Oh! Whoops.
>
>>>>I also wonder about numerical accuracy.
>>>
>>>Do we expect x86 numerics to be *WORSE* than GPU numerics?
>>
>>No, but I suspect the SNB GPU may have worse numerics than Nvidia/AMD GPUs.
>
>This would not surprise me, either.
>
>>Agreed. Although I think Matt Pharr and the other folks in ART are trying to give
>>you *similar* performance, without using instrinsics.
>
>Folks are trying. I've spent time in the last few years dealing with nVidia/CUDA
>claims. We've coded for nVidia/CUDA, ATI/OpenCL and evaluated RapidMind (now Intel
>Array Building Blocks), and Intel Thread Building Blocks on x86.
>
>None of them come anywhere close to their performance claims on our loads (which,
>I grant, are not single precision floating point ...).
>
>We couldn't even get ABB to beat ICC compiling *scalar* code and auto-vectorization
>turned on, although ABB beat GCC by a few tens of percent. RapidMind failed just
>as badly a few years ago when we evaluated them and asked *them* to write the RapidMind code for us.
>
>So ... I'm skeptical about OpenCL on x86 beating hand >coded intrinsics in general any time soon.

I remember our discussions : ) BTW - are you guys going to Hot Chips this year?

You might want to try ISPC. I'd be very curious to know how it stacks up against well coded software (http://ispc.github.com/).

David
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Sandy Bridge GPU article onlineDavid Kanter2011/08/09 03:53 AM
  Sandy Bridge GPU article onlinesJ2011/08/09 07:29 AM
    Sandy Bridge GPU article onlineDavid Kanter2011/08/09 12:11 PM
  Sandy Bridge GPU article onlineKevin G2011/08/09 07:32 AM
    Sandy Bridge GPU article onlineDavid Kanter2011/08/09 12:14 PM
      Sandy Bridge GPU article onlineanon2011/08/09 11:43 PM
  Sandy Bridge GPU article onlineJosh2011/08/09 09:39 AM
    Sandy Bridge GPU article onlineMr. Camel2011/08/09 12:39 PM
      Sandy Bridge GPU article onlineJosh2011/08/09 12:48 PM
        Sandy Bridge GPU article onlineDavid Kanter2011/08/09 01:06 PM
          Sandy Bridge GPU article onlineJosh2011/08/09 01:18 PM
            Sandy Bridge GPU article onlineDavid Kanter2011/08/09 01:50 PM
              Sandy Bridge GPU article onlineJosh2011/08/09 02:22 PM
                Product release milestonesDavid Kanter2011/08/10 10:42 AM
      Sandy Bridge GPU article onlineSeni2011/08/09 02:27 PM
    Sandy Bridge GPU article onlineDavid Kanter2011/08/09 01:12 PM
  Sandy Bridge GPU article onlinerscheidegger2011/08/09 01:27 PM
    Sandy Bridge GPU article onlineDavid Kanter2011/08/09 01:45 PM
      Sandy Bridge GPU article onlinerscheidegger2011/08/09 04:34 PM
      Sandy Bridge GPU article onlinerscheidegger2011/08/09 06:03 PM
  Sandy Bridge GPU article onlineSi2011/08/10 05:15 AM
  Some error on the chart?Ron2011/08/10 09:33 AM
    Some error on the chart?David Kanter2011/08/10 10:35 AM
  Sandy Bridge GPU article onlineGroo2011/08/10 10:01 AM
    Sandy Bridge GPU article onlineArcadian2011/08/10 02:38 PM
      Sandy Bridge GPU article onlineGroo2011/08/10 08:32 PM
        Sandy Bridge GPU article onlineIntelUser20002011/08/11 03:17 AM
          Sandy Bridge GPU article onlinePoindexter2011/08/11 04:51 AM
            Sandy Bridge GPU article onlineIntelUser20002011/08/11 08:32 AM
              Sandy Bridge GPU article onlineGroo2011/08/11 12:58 PM
                Sandy Bridge GPU article onlineDavid Kanter2011/08/12 03:16 PM
                  Sandy Bridge GPU article onlineGroo2011/08/15 08:09 PM
                    Sandy Bridge GPU article onlineDavid Kanter2011/08/16 01:00 AM
        Sandy Bridge GPU article onlineLinus Torvalds2011/08/11 11:20 AM
          Sandy Bridge GPU article onlineGroo2011/08/11 12:51 PM
        AMD could be worse.Jouni Osmala2011/08/11 11:50 PM
          AMD could be worse.bakaneko2011/08/12 01:53 AM
  OpenCL, Compute Shader 4.x Support, and OpenGL 3.1 Supportltcommander.data2011/08/10 03:02 PM
    OpenCL, Compute Shader 4.x Support, and OpenGL 3.1 SupportDavid Kanter2011/08/10 06:01 PM
      OpenCL should be able to run on x86 ... but probably not fastMark Roulo2011/08/15 03:30 PM
        OpenCL should be able to run on x86 ... but probably not fastRohit2011/08/15 06:52 PM
        OpenCL should be able to run on x86 ... but probably not fastDavid Kanter2011/08/15 07:24 PM
          OpenCL should be able to run on x86 ... but probably not fastMark Roulo2011/08/16 06:47 AM
            OpenCL should be able to run on x86 ... but probably not fastDavid Kanter2011/08/16 07:49 AM
        OpenCL should be able to run on x86 ... but probably not fastSylvain Collange2011/08/16 03:04 AM
    OpenCL, Compute Shader 4.x Support, and OpenGL 3.1 SupportSylvain Collange2011/08/13 07:54 AM
  How much raw computer resources?Robert Davide Graham2011/08/12 12:02 PM
    How much raw computer resources?David Kanter2011/08/12 12:25 PM
      How much raw computer resources?Robert David Graham2011/08/12 07:44 PM
        How much raw computer resources?Nicolas Capens2011/08/14 11:51 PM
          How much raw computer resources?Robert David Graham2011/08/15 11:54 AM
            How much raw computer resources?Michael S2011/08/15 03:08 PM
            How much raw computer resources?David Kanter2011/08/15 07:28 PM
              Bitcoin miner: a benchmarkRobert Daivid Graham2011/08/16 04:03 PM
                Bitcoin miner: a benchmark?David Kanter2011/08/16 04:31 PM
                  Buying with bitcoinsRohit2011/08/16 07:11 PM
                    Buying with bitcoinsDavid Kanter2011/08/16 08:34 PM
                      Buying with bitcoinsRobert David Graham2011/08/17 06:47 PM
                  aside: bitcoinhobold2011/08/17 03:47 AM
                  Bitcoin miner: a benchmark?Max2011/08/18 03:54 AM
                  Bitcoin miner: a benchmark?etzel2011/08/18 07:52 AM
    How much raw computer resources?Nicolas Capens2011/08/14 12:04 PM
  I hate to be the text critic...A Hobbit2011/08/19 12:59 AM
    You're rightDavid Kanter2011/08/19 09:16 PM
      Fixed (NT)David Kanter2011/08/19 09:30 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?