GPCBenchmark - A OpenCL General Purpose Computing benchmark

Discussion in 'GPGPU Technology & Programming' started by Arnold Beckenbauer, Apr 30, 2010.

  1. Arnold Beckenbauer

    Veteran

    Joined:
    Oct 11, 2006
    Messages:
    1,357
    Likes Received:
    298
    Location:
    Germany
    http://www.hpctech.com/down/opencl/index_en.html

    Opinions? It shows that my HD4850 isn't good for OpenCL.
     
  2. CarstenS

    Veteran Subscriber

    Joined:
    May 31, 2002
    Messages:
    4,681
    Likes Received:
    1,885
    Location:
    Germany
    From what I'm seeing, a potent GPU dwarfs a CPU in every respect of theoretical throughputs by several orders of magnitude. This benchmark reinforces that preconception.

    - Nice to see DPFP included - Radeons should wipe the floor with consumer level Geforces here
    - Common Maths gives up on me - no results after several minutes.
     
    #2 CarstenS, Apr 30, 2010
    Last edited by a moderator: Apr 30, 2010
  3. rpg.314

    Veteran

    Joined:
    Jul 21, 2008
    Messages:
    4,298
    Likes Received:
    0
    Location:
    /
    Only in the right workloads.
     
  4. CarstenS

    Veteran Subscriber

    Joined:
    May 31, 2002
    Messages:
    4,681
    Likes Received:
    1,885
    Location:
    Germany
    Please see my edit in italics. :)
     
  5. fellix

    fellix Hey, You!
    Veteran

    Joined:
    Dec 4, 2004
    Messages:
    3,423
    Likes Received:
    301
    Location:
    Varna, Bulgaria
    #5 fellix, Apr 30, 2010
    Last edited by a moderator: Apr 30, 2010
  6. Arnold Beckenbauer

    Veteran

    Joined:
    Oct 11, 2006
    Messages:
    1,357
    Likes Received:
    298
    Location:
    Germany
  7. fellix

    fellix Hey, You!
    Veteran

    Joined:
    Dec 4, 2004
    Messages:
    3,423
    Likes Received:
    301
    Location:
    Varna, Bulgaria
    Any numbers from Fermi?
     
  8. CarstenS

    Veteran Subscriber

    Joined:
    May 31, 2002
    Messages:
    4,681
    Likes Received:
    1,885
    Location:
    Germany
    Sorry for the inconvenient format, i only have it inside a spreadsheet here.

    GTX 480

    105,697
    147,146
    136,741
    116,503
    6,162
    6,642

    1307,222
    1301,323

    578,345
    288,979
    587,512
    495,233

    630,602
    630,459
    1259,871
    21,639
    65,519

    82,348
    82,35
    164,998
    4,717

    3133,332
    1396,644
    5914,001
    5606,886
    3754,879
    3983,106
    125,824
    69,67

    NA
    1489,157
    929,233
    3836,429
    NA
    940,939
    1228,487
    NA
    130,829
    232,868
    NA
    1278,363
    3185,012

    409,727
     
  9. fellix

    fellix Hey, You!
    Veteran

    Joined:
    Dec 4, 2004
    Messages:
    3,423
    Likes Received:
    301
    Location:
    Varna, Bulgaria
    Double Ops is working now (added to the score list), looks like the program doesn't like to be overloaded with continuous test runs, sort of. ;)
     
  10. Arnold Beckenbauer

    Veteran

    Joined:
    Oct 11, 2006
    Messages:
    1,357
    Likes Received:
    298
    Location:
    Germany
    The DP-MADD result is really high.
     
  11. Florin

    Florin Merrily dodgy
    Veteran

    Joined:
    Aug 27, 2003
    Messages:
    1,601
    Likes Received:
    133
    Location:
    The colonies
    Overclocked too far?
     
  12. CarstenS

    Veteran Subscriber

    Joined:
    May 31, 2002
    Messages:
    4,681
    Likes Received:
    1,885
    Location:
    Germany
    It's actually what you'd expect from 1.400 MHz, 2 FLOPS per Cycle and 64 DFMAs/clk.


    edit:
    On an additional note: I'm running the HD 5870 in OpenCL now and of course the Stream SDK installed the CPU drivers for OpenCL also. With this driver the systems Core i7 is about one half to one tenth as fast as without. WTF?
     
    #12 CarstenS, Apr 30, 2010
    Last edited by a moderator: Apr 30, 2010
  13. Jawed

    Legend

    Joined:
    Oct 2, 2004
    Messages:
    10,853
    Likes Received:
    722
    Location:
    London
    Nice set of basics here (though multiplying 256x256 matrices is hardly a test, that kind of size hardly wakes up a big GPU). They'll provide a nice barometer as AMD sorts out OpenCL...

    NVidia's "application" performance looks much healthier.

    Jawed
     
  14. CarstenS

    Veteran Subscriber

    Joined:
    May 31, 2002
    Messages:
    4,681
    Likes Received:
    1,885
    Location:
    Germany
    Right,in this benchmark SP Matrix Multiply for example is a case, where Cypress cannot be put to use - Juniper is only 20% slower there.
     
  15. Florin

    Florin Merrily dodgy
    Veteran

    Joined:
    Aug 27, 2003
    Messages:
    1,601
    Likes Received:
    133
    Location:
    The colonies
    The Fermi is in my second PC at the moment while I'm having an Eyefinity weekend so check some low end CPU scores here.

    BTW if Carsten's scores are at GPU stock speeds like mine then there seems to be a bit of CPU dependency to the GPU scores, as mine are generally a tad lower (this is driver 197.41 on Windows 7 64-bit, ATI Stream SDK 2.01 also installed btw).

    Any 5870 stock scores out there?

    Device Info page

    Global Memory
    Local Memory
    Int32 Ops
    Float Ops
    Double Ops
    Common Math
    Image Processing
    Cryption
     
  16. Albuquerque

    Albuquerque Red-headed step child
    Veteran

    Joined:
    Jun 17, 2004
    Messages:
    3,845
    Likes Received:
    329
    Location:
    35.1415,-90.056
    From the other cypress pictures, there must be a newer OpenCL SDK that I don't have yet. I'll go find it, install it, and post benches of my 5850 at stock and at 950/1150 on my C2Q 9550 @ 3.73GHz
     
  17. Jawed

    Legend

    Joined:
    Oct 2, 2004
    Messages:
    10,853
    Likes Received:
    722
    Location:
    London
    The graphics driver contains the ultimate compiler for GPU execution, e.g. in ATI the IL generated by OpenCL compilation has to be translated into ISA. So for ATI results are a combination of SDK and Catalyst.
     
  18. Arnold Beckenbauer

    Veteran

    Joined:
    Oct 11, 2006
    Messages:
    1,357
    Likes Received:
    298
    Location:
    Germany
  19. Dade

    Newcomer

    Joined:
    Dec 20, 2009
    Messages:
    206
    Likes Received:
    20
  20. Jawed

    Legend

    Joined:
    Oct 2, 2004
    Messages:
    10,853
    Likes Received:
    722
    Location:
    London

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...