FluidMark 1.3.0: GPU PhysX Fluids with more than 700K SPH Particles (GPU Tool, Updated)

FluidMark 1.3.0 - PhysX benchmark

FluidMark 1.3.1 DOWNLOAD

You can download PhysX FluidMark 1.3.1 here:
Webmasters: hotlinking is not allowed, please use the post url as download link.

FluidMark 1.3.0 release highlights

The new version of FluidMark is available. This new version 1.3.0 fixes an important bug related to GPU memory allocation and now you can set a massive number of particles in GPU PhysX: more than 700’000. Actually the maximum number of particles is around 779’000 (PhysX hardware limitation and / or PhysX driver…). But this is not really important because you have to have a very high end system to run 700K SPH particles (GPU or CPU).

PhysX FluidMark - 700K particles on a GTX 480
700’000 PhysX particles on a GTX 480

PhysX FluidMark - 600K particles on a GTX 480
600’000 PhysX particles on a GTX 480

Another important change concerns the scoring system. Now, the global score is the sum of PhysX score and 3D score. This simple sum clearly shows the gain brought by a dedicated PhysX GPU.

Here are two scores with the following settings: 120’000 particles, 7 emitters, 1920×1080 fullscreen, PostFX enabled, Async mode ON:

  • GTX 480 for PhysX and 3D – Global score: 548 points, PhysX score: 138 points (23 SPS avg), GraphX score: 409 points (67 AVG avg)
  • GTX 480 for 3D and GT 240 for PhysX – Global score: 623 points, PhysX score: 57 points (9 SPS avg), GraphX score: 566 points (93 AVG avg)

As you can see, the GT 240 brings a non negligible contribution in the global score.

The scores page is available here: FluidMark 1.3.x Scores Page.

I also added an additional graphics load with the More graphics load checkbox. When enabled, this adds a pure pixel shader load on the GPU by rendering a quad on the top-right corner:

PhysX FluidMark - 600K particles on a GTX 480
The additional graphics load in the top-righ corner.

I took this pixel shader from iq’s Shader Toy, a great webapp for testing WebGL shaders. Shader credits:

‘Nautilus’ by Weyland Yutani (reworked by iq) (2010)
A 1 kilobyte demo from the demoscene, using raymarching and screen space ambient occlusion.

What is FluidMark?

FluidMark is a fluid simulation benchmark based on NVIDIA PhysX engine. The fluid simulation uses the SPH or Smoothed Particle Hydrodynamics method where interparticle forces are considered (SPH requires more CPU or GPU horsepower!). FluidMark support both CPU and GPU PhysX and in CPU mode, multi-core CPUs are supported (see HERE).

NVIDIA PhysX logo

FluidMark 1.3.1 changelog

  • Bugfix: online submission didn’t work properly when the nickname field was filled.
  • Change: now Win32 Setup installs two shortcuts: FluidMark.exe (single GPU) and etqw.exe (SLI / CrossFire).

FluidMark 1.3.0 changelog

  • New: scoring system change. Now the global score is the sum of PhysX score and GraphX (3D rendering) score. The global score is more appropriate to reflect the gain brought by a dedicated PhysX card.
  • New: added a new option to increase the graphics workload ([More graphics load] checkbox).
  • Bugfix: fixed GPU memory allocation. Now you can run more than 700’000 SPH particles.
  • Change: online submission improved.
  • Change: updated GPU monitoring code with latest version of ZoomGPU (1.5.11)
  • Change: removed [Heavy PhysX mode] and [Use particle count] checkboxes.

35 thoughts on “FluidMark 1.3.0: GPU PhysX Fluids with more than 700K SPH Particles (GPU Tool, Updated)”

  1. WacKEDmaN

    awesome work JeGX!
    looks like you out did yaself yet again! 🙂
    nice job on fixing the particle count bug

  2. WacKEDmaN

    hmm..im having an issue with my 9400GT as a secondary card, i have it selected in nvcontrol panel, but 1.3.0 keeps using CPU physx.

    1.3.0 is using my GTX470 fine when its selected in nvcpl

    i tested against fluidmark 1.2.2 and it is still working as expected..

  3. JeGX Post Author

    Thanks for this feedback. I think that the 9400 GT doesn’t like the new GPU memory allocation technique that’s why you get CPU PhysX. I will look at this bug asap.

  4. Rik55

    I also can’t get it to work.
    HD4850 on a core2duo. CPU cores go to 100% and slows system to a halt.
    1.2.2 still works normaly
    Running HD4850. Core2Duo6300.
    Cat10.10e 8.782.1, PhysX 9100514

  5. Rik55

    Got it working. Not to sure what happened. I first had to actually Force PhysX on the CPU for it to work on ATI hardware. But now it works whether PhysX CPU is ticked or not. ???

  6. Rik55

    back again. It seems it was the Multi GPU version (etqw.exe) that is locking up on my single ATI 4850 GPU. CPU shoots straight up to 95-100% before I even run any test. The normal Fluidmark.exe works fine.

  7. WacKEDmaN

    @JeGX: yup, i did end up with a memory failure log at one stage..

    GpuHeapMalloc failure of shapeImpulseOrNormalArray in file ‘d:\sw\physx\physxsdk\2.8.4\trunk\lowlevel\gpu\common\include\PxgFluidParticleCollisionData.h’ in line 119.

    but im having another issue aswell with the submitting of benchmark scores. it seems i can only submit my name when POSTFX is DISABLED! (and the other submission requirements are met)..its inflating my score by about 100 points! 606 points for a 470!

  8. JeGX Post Author

    @Rick55: for the next update, I will put shortcuts to both exe: FluidMark.exe for single GPU and etqw.exe for multi-GPUs. But your problem is strange because I tested FluidMark on a HD 6870 + Cat 10.11 without hitch. I’ll test fluidmark with a HD 4850.

    @WacKEDmaN: yep there’s a bug. I’m updating FluidMark…

  9. JeGX Post Author

    @WacKEDmaN: there is a command line param for the GPU memory allocation: /physx_gpu_heap_size

    Try different values, for example 128 or 256:

    FluidMark.exe /physx_gpu_heap_size=128

    The default value is 512.

  10. Stefan

    Some issues with ATI 5470/Intel GMA HD hybrid graphics notebook:

    POSTFX shader creation fails with Intel GMA HD renderer (any Fluidmark version), “more graphics load” works fine though.

    If “PhysX CPU” is ticked, START takes ages and the notebook’s “switchable graphics” control panel goes nuts.
    Benchmark results in diff. scores:
    “PhysX CPU” is ticked: 18/3/21
    “PhysX CPU” not ticked: 15/14/30

  11. Stefan

    It appears that start delay was caused by renaming, so i did a row of tests:

    Catalyst 10.11, 640×480, PhysX/GFX/Points
    fluidmark.exe 19/107/126
    etqw.exe 13/23/37
    Heaven.exe 30/1/31

    Catalyst 10.10E, 640×480, PhysX/GFX/Points
    fluidmark.exe 21/79/100
    etqw.exe 0/0/2 !
    Heaven.exe 18/184/203

  12. Jason

    It seems GPU physx doesn’t work with the ati hybrid physx patch. I get 0% usage on my GT 240. Previous version worked fine.

  13. Rik55

    @Stefan @JeGX
    Yip, after looking into the renaming of files, removing any Catalyst Application Profiles (CAP) or simply disabled Catalyst AI to remove any ATI optimisations allow both etqw.exe and fluidmark.exe to work as normal with the same scores.

  14. Stefan

    OK, no more start delay after disabling Cat AI – i have no CAP installed – yet diff. results:

    Catalyst 10.10E, 640×480, PhysX/GFX/Points
    fluidmark.exe 17/94/111
    etqw.exe 27/9/36
    Heaven.exe 18/58/77

    Another thing:
    I’d like an option to disable warming-up, it takes 150 seconds on my notebook.

  15. WacKEDmaN

    Thank you JeGX..256 setting worked a charm, i tryed with 128 first up, but particles disapeared after 64K..

    new score of 719 with both cards (and no postfx!)

  16. WacKEDmaN

    just an after thought on the benchmark submission, is it possable to get the 2nd cards info submitted? maybe only when using the multi-gpu version…as it is, just showing the first card defeats the purpose when a dedicated card is being used as the scores are generally higher but it looks like the primary card was doing the work

  17. WacKEDmaN

    @stefan: warmup is needed for the benchmark, it loads the total particle count before actually starting the benchmark proper. you get higher fps when it first starts emmitting the particles, causing the score would be inflated as you would never get all the particles loaded

    if u need it disabled, use stability test maybe?

  18. Stefan

    You’re right. I use lower particle count now for testing the renaming thing etc.

    Out of curiousity i tried it the other way round.
    Particle count: 2^19=524288 (default heap), time-base: 20 minutes
    After 10 minutes warming up stops with ~412000 particles and then reaches maximum=454545 after a few minutes.

  19. WacKEDmaN

    some interesting results on my system…

    3D/GTX470 + Physx/9400GT:
    Global score: 597 points
    PhysX score: 27 points (4 SPS avg)
    GraphX score: 569 points (93 AVG avg)

    Global score: 514 points
    PhysX score: 130 points (21 SPS avg)
    GraphX score: 384 points (63 AVG avg)

    3D/GTX470 + CPU Physx/Q8200@2.5Ghz
    Global score: 508 points
    PhysX score: 20 points (3 SPS avg)
    GraphX score: 487 points (80 AVG avg)

    as you can see..the 9400GT is only slightly faster than the Q2D CPU..but the cpu score doesnt come close..due to the bottlenecking from the cpu running at 100% on all cores..even seemed to case a random studdering during the benchmark, even the the fps remained high

  20. Pingback: Hybrid PhysX mod v1.03 / v1.04ff - Page 100

  21. JEskandari

    I have a small problem with my Palit GTX460 Sonic
    and this program .
    it run well and have no problem but when I want to exit the program the program hangs and I have terminate the Process to exit it.
    also it happen before it give a score in benchmark mode so I can’t receive any score.
    right now I use Windows 7 X64 and Geforce 263.06

  22. Pingback: PhysX FluidMark 1.3.1 Available For Download - Expreview.com

  23. Pingback: Tools - Page 33

  24. JEskandari

    well I did as you mentioned but still I have the Same problem
    by the way thanks for the replay

  25. Hornet

    Hello, I’m having some issue here where it doesn’t seems to be using my dedicated Physx card.

    My setup is a GeForce GTX 295 as the main card, and a GeForce 9800GT as a dedicated Physx card. In the control panel, I’ve set the 9800GT as a dedicated Physx card

    No matter what setting I use, it either use the CPU to do the physics workload, or the GTX 295 itself. Thanks

  26. Razorwindmo46

    I am having the same issues weather using 1.30 or 1.31 after renaming or deleting all Physx.dll files
    Physics warmup loads then after a few seconds particles disappear.

  27. WacKEDmaN

    @Hornet: try running fluidmark with the command line paramater..

    FluidMark.exe /physx_gpu_heap_size=256

    if that doesnt work try =128

    @JeGX: looks like the heap size of 512 might be a bit too much..especially for cards with only 512mb!..256 works great on my 9400

  28. Sparky78

    Hi, I’m also having a problem a with hybrid system;2 x XFX 5850HD and XFX 240GT. The 240GT isn’t used. I have tried the heap size with no difference. Version 1.2.0 still works.

  29. Sparky78

    Possible fix, if I rename PhysXDevice.dll then PhysX uses the GPU.

  30. Andrey

    i havent gpu acceleration on my 9600gso
    in control panel physx on gpu.but in fluidmark runs on cpu. wtf?

  31. sendblink23

    I experience the same exact issue as this other user commented previously on 1.3.1 multi-gpu:

    “Rik55 says:
    November 26, 2010 at 8:36 pm
    back again. It seems it was the Multi GPU version (etqw.exe) that is locking up on my single ATI 4850 GPU. CPU shoots straight up to 95-100% before I even run any test. The normal Fluidmark.exe works fine.”

    I use CrossfireX 5770’s + 9800GTX+ – the multi-gpu version without even running a test the CPU usage on my AMD 965 skyrockets to full 100% – and when I start the test my system locks up like a snail… I have to Ctrl + Alt + Delete to kill the process… and restart the computer because it stays 100% usage even after killing it.

    There is clearly a very odd bug on the multi gpu version.

  32. Zibri

    434 points with dua GTX 460 doesn’t seem right.
    One card had 100% cpu the other one: 0

    What’s up?

  33. Misted.Mind

    Fluidmark wont run for me on GPU no matter what i change… It always lets the particles disappear after about half a second when in stability mode. IN benchmark-mode it will crash the driver or remove the particles only…
    I reduced the particle count to like 64k, or even 10k, with 3 emitters…
    Or raise them to 180k, or 200k with 10 emitters… It will only work in crappy slow CPU-mode…

    Man do i hate when this… It works perfectly at my brothers GTX470.
    But not with my GTX260…
    Same driver version 266.58 physX driver I tried 9100513 & 9100514…

  34. Gabkicks

    was there ever a fix found for hybrid physix and fluidmark

Comments are closed.