DirectCompute (Direct3D11 Compute Shaders)

Discussion in 'GPGPU Technology & Programming' started by Scali, Sep 11, 2009.

  1. Jawed

    Legend

    Joined:
    Oct 2, 2004
    Messages:
    10,873
    Likes Received:
    767
    Location:
    London
    I wonder if DP uses more area per SIMD in R800? It's supposedly been improved.

    Jawed
     
  2. trinibwoy

    trinibwoy Meh
    Legend

    Joined:
    Mar 17, 2004
    Messages:
    10,436
    Likes Received:
    442
    Location:
    New York
    Why would it? DirectCompute is still targeted at games and consumer apps, not HPC.
     
  3. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    Not a mistake, but a by-product of its development path.

    Indeed, as are mainstream and entry level GPU's.
     
  4. Mintmaster

    Veteran

    Joined:
    Mar 31, 2002
    Messages:
    3,897
    Likes Received:
    87
    It's possible. Devs just had to get the go-ahead to make a DX9 path from their superiors, and there's nothing like a marketshare number (real or projected) to make your case.

    On a related note, sometimes eagerness to support the next gen was actually detrimental. 3DMark2001, for example, was so focussed on using DX8 pixel shaders that they used the texm3x3vspec instruction instead of EMBM, which is not only a DX7 feature but also gives much more accurate reflections for planar surfaces. This happened in both the pixel shader test and the Nature test, along with most DX8 games.

    True, but it doesn't matter what the user experience is. The decision to use DX9 starts way before then, possibly before NVidia even launched the Geforce FX.

    I'm not sure if the rudimentary DP in RV670/RV7xx was costly enough to warrant a design change. There's probably other motives, too, as the 5850 is cheap enough that cost is not going to be a barrier for anyone interested in making real apps that need DP calcs, so might as well take away the lower cost option for this niche market.
     
  5. Mendel

    Mendel Mr. Upgrade
    Veteran

    Joined:
    Nov 28, 2003
    Messages:
    1,350
    Likes Received:
    17
    Location:
    Finland
  6. Florin

    Florin Merrily dodgy
    Veteran Subscriber

    Joined:
    Aug 27, 2003
    Messages:
    1,648
    Likes Received:
    219
    Location:
    The colonies
    k..

    DirectCompute Benchmark v0.15

    C2D E6750 with GTX280 (stock)

    Windows 7:
    GPU Score M16591/S2992
    CPU Score M2642/S1321

    Vista:
    GPU Score M16664/S2636
    CPU Score M2882/S1324

    As a side note, unlike F@H, DC Mandelbrot, Furmark etc, this GPU doesn't really sound busy during this bench - if you know what I mean..
     
  7. mikegi

    Newcomer

    Joined:
    Nov 20, 2006
    Messages:
    27
    Likes Received:
    2
    Those GPU scores don't make any sense. If "M" means multithreaded and "S" means single-threaded, then the two scores should be about equal for the GPU assuming the only difference in the test is the thread count. Instead, the single-threaded scores are about the same as the MT CPU score.
     
  8. Broken Hope

    Regular

    Joined:
    Jul 13, 2004
    Messages:
    483
    Likes Received:
    1
    Location:
    England
    Here's my results

    Q9650 with a 5870 using the 9.11 beta drivers.

    [​IMG]

    [​IMG]
     
  9. EduardoS

    Newcomer

    Joined:
    Nov 8, 2008
    Messages:
    131
    Likes Received:
    0
    What the **** is the GPU multi-threaded test?
     
  10. Mendel

    Mendel Mr. Upgrade
    Veteran

    Joined:
    Nov 28, 2003
    Messages:
    1,350
    Likes Received:
    17
    Location:
    Finland
    My 5870 results are very very close to Broken Hope's.

    I'm mightily puzzled by Florin's scores though.
     
  11. Psycho

    Regular

    Joined:
    Jun 7, 2008
    Messages:
    745
    Likes Received:
    39
    Location:
    Copenhagen
    Strange results.. 5770@stock / E6750@3.3, cat9.11 opencl beta:
    GPU: M33922/S33916
    CPU: M602/S309
    5770@960/1200: M37907
    5770@960/1440: M38324

    I wonder if the 5800s didn't clock up... And generally I don't find a benchmark doing WHAT? particulary relevant - got no idea what's really inside and what it's optimized against.
     
    #71 Psycho, Nov 7, 2009
    Last edited by a moderator: Nov 7, 2009
  12. Broken Hope

    Regular

    Joined:
    Jul 13, 2004
    Messages:
    483
    Likes Received:
    1
    Location:
    England
    Looks like there is a new version, I get numbers that are closer to Psycho's now.

    [​IMG]
     
  13. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    8,254
    Likes Received:
    1,940
    Location:
    Finland
    That's still ~half of what it should be, as Psycho has 5700-series
     
  14. Broken Hope

    Regular

    Joined:
    Jul 13, 2004
    Messages:
    483
    Likes Received:
    1
    Location:
    England
    Heh just noticed that, the benchmark doesn't make much sense at all.
     
  15. Broken Hope

    Regular

    Joined:
    Jul 13, 2004
    Messages:
    483
    Likes Received:
    1
    Location:
    England
    Yep something not right with this benchmark, this is from another forum I visit.

    [​IMG]

    They also get higher CPU score even though they are using a dual core CPU where's I'm using a quad core.
     
  16. EduardoS

    Newcomer

    Joined:
    Nov 8, 2008
    Messages:
    131
    Likes Received:
    0
    MScore = (Vendor == 0x10DE) ? SScore * multiplier : SScore; :D
     
  17. Psycho

    Regular

    Joined:
    Jun 7, 2008
    Messages:
    745
    Likes Received:
    39
    Location:
    Copenhagen
    yeah.. that mscore/sscore difference for the geforces doesn't make sense either; if it was really a gpu test it shouldn't matter if you only communicate with dx from a single thread..
    I think it's best just to forget about this "benchmark"...
     
  18. Regeneration

    Newcomer

    Joined:
    Sep 17, 2005
    Messages:
    37
    Likes Received:
    1
    Before you rush to cry foul, please take in mind that some drivers aren't mature enough.
     
  19. Broken Hope

    Regular

    Joined:
    Jul 13, 2004
    Messages:
    483
    Likes Received:
    1
    Location:
    England
    I can't see how immature drivers would make a 5770 score around the same as a 5870 when the 5770 has half the number of shaders and only a 128 bit bus.

    There's something very odd with the benchmark.
     
  20. Regeneration

    Newcomer

    Joined:
    Sep 17, 2005
    Messages:
    37
    Likes Received:
    1
    Rendering is not the same as computing.
     
Loading...

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...