ComputeMark 1.3 Available and GTX 470 and HD 5870 Scores

ComputeMark - DX11 DirectCompute benchmark

The new version of ComputeMark is available. ComputeMark is a GPU computing benchmark based on Microsoft DirectX 11 DirectCompute technology.

You can download the latest version here:

This benchmark loves Radeon HD 5000 series…

Here are some results on an Intel Core2 E8400 CPU @ 3GHz, 2GB DDR3, Windows 7 64-bit and Catalyst 10.4 / ForceWare 197.75:

ComputeMark - DX11 DirectCompute benchmark - GeForce GTX 470
GeForce GTX 470 (forceware 197.75) score: 589

ComputeMark - DX11 DirectCompute benchmark - Radeon HD 5870
Radeon HD 5870 (Catalyst 10.4) score: 2374

ComputeMark - DX11 DirectCompute benchmark - Radeon HD 5770
Radeon HD 5770 (Catalyst 10.4) score: 1222


23 thoughts on “ComputeMark 1.3 Available and GTX 470 and HD 5870 Scores”

  1. truth

    nvidia gtx 470 = 589,

    ati hd 5870 = 2374!!!

    the score difference is amazing…goodjob ati. i wish to see the physx thing in ati.

  2. Groovounet

    Erm… I would like to see what kind of computation is done. A Radeon 5770 twice the score of a GeForce 470 is just non-sense.

  3. WacKEDmaN

    yeah identical score on my GTX470..

    seems its optimised for ATI

  4. Grandmaster

    Version 1.3 has optimizations for ATI and Nvidia. Good job ATI!

  5. Psolord

    Here’s my result on my 5850s crossfire, clocked at 1Ghz each!

    The score is 4469! lol

    The benchmark is NOT optimised for ATI. I had the exact same question, which I asked Robert himself in VrZone and he doesn’t really know why ATI performs so much better. One possible reason was Nvidia’s drivers.

  6. jj99

    HD 5870 is 2.5x the raw power of GTX 470. Fermi is optimized for complicated execution control, not plain arithmetics like in this benchmark. So in fact, it’s difficult to optimize the program for Radeon, to make it simple enough, so all the raw power is used. Also Radeon is optimized for vector arithmetic, but Fermi can work with scalars too. So if the code is good enough for Radeon, it must be good for GeForce too. Still this don’t explain 4x difference… Maybe just drivers are not mature enough…

  7. Joshua

    I ran the bench on my GTX480 this morning. My score, with 180 sec, is 4307.

  8. 008

    Maybe it’s because of the sacristy in the number of Shader units.

    5770 – 800 unified
    GTX 480 – 480 u

  9. Ourasi

    “Joshua says:
    May 11, 2010 at 2:48 pm

    I ran the bench on my GTX480 this morning. My score, with 180 sec, is 4307.”

    Score using 180s Burn on my HD5970@850/1200 is 28461, so that burn option, since being so much longer, naturally gives higher scores, and widens the gap even further it seems.

    Benchmark run gives me 4404..

  10. sgtheadhole

    have you guys seen the SP performance of ATi cards, they woop nvidias cards ass.

    5870=2.7 TFlops
    5770=1.36 TFlops
    GTX470=1 TFlop

  11. Remi

    “Version 1.3 has optimizations for ATI and Nvidia.”
    Sorry, but when it comes to compute, that just doesn’t make sense.
    Either you don’t optimize, or you optimize for one architecture, but you can’t optimizing for both. Optimizing for ATI implies writing vectorized code while optimizing for nVidia implies writing scalar code. As they are contradictory, you just can’t have both.

  12. DrBalthar

    Well if you follow some discussion I think one of the authors Robert Varga was actually contacted by nVidia devteam and the crap performance on nVidia is simply down to the memory model used (3D textures). It is not the first time I see sloopy implementation on certain texture formats on nVidia hardware.

  13. Remi

    Ok… IIRC they recommend using a 2D texture divided into slices to implement a 3D texture (cause it’ll perform better). Once it’ll been taken care of, the benchmark’s results might have a chance to reflect the performance of “real life application”. 🙂 Until then, the benchmark doesn’t really live up to its promise, I’m afraid. Too bad.

  14. Fool Gamer

    compare 5870 and 480

    ati have very good hardware development (low TDP and good)
    and poor software support(ati stream)
    (because of poor software support ati hardware did’t come to show the true high perfomance like 5970)

    nvidia have good hardware development (High TDP and high perfomance)
    and rich software support (Nvidia cuda+physic+OpenGl)

  15. DrBalthar

    Wrong 3D textures makes extremly more sense since the fluid it computes is an actual volume instead of mangling it down in an awkward 2D Texture Array. Who’s to blame is definitely nVidia not the authors of ComputeMark.

  16. John


    Sorry but it is not the fault of the benchmark if it uses the most perfectly suitable datatype for fluid dynamics just because one company failed to correctly implement support for it in their latest hardware. What you’re proposing isn’t a benchmark–it’s a demo.

  17. Joshua

    The proof is in the games. Not in these pathetic little benchmark programs !

  18. Psolord

    Correct. The proof is in games. The GTX 480 is faster than the 5870 period.

    Oh wait a minute. It is only 15% faster, yet it consumes 50% more power in load and like 300% more in idle, while it costs 50% more as well. WTF?

  19. Fool Gamer

    Correct. The proof is in games
    yup that right

    amd/nvidia please give us a good vga driver
    to fight all pc game release this year 2010


  20. Bobo


    Using V1.3 Score for Burn Test: 4204
    Benchmark 696

    AMD PhenomII 955 BE @3.4GHz for today
    ATI OGL4March29 w/10.4 WHQL but, uninstalled 10.4 Display Drivers and installed OGL4 Display Drivers. What a difference in FPS! Hey! ATI Kicks Nvidia out of town… boom, whop, splat..

Comments are closed.