Announcement

Collapse
No announcement yet.

V-Ray GPU vs FStorm Renderspeed

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • V-Ray GPU vs FStorm Renderspeed

    Hi,

    I am trying to compare FStorm with V-Ray. That's why I build a simple scene using the exactly same ressources and a get a pretty matching result and used very similar shaders.
    The weird thing is that FStorm takes just a fraction of the time compared to V-Ray to clear up the image. For V-Ray to reach a similar result than F-Storm it would probably take 5 or more times the amount. Even a very simple grey diffuse material (no reflections) takes a long time and produces very visible noise.
    Hardware is the same (RTX 4090), no Denoiser, no render elements.

    I understand both are different engines and V-Ray has a much more complete featureset but how is it possible that there is this massive difference in rendertime/quality?
    I prefere the visual output of the V-Ray a little bit more, where it feels like the colors are bouncing around a little bit more and make the shadows in general richer. But some of that I think much comes down to tonemapping differences.

    V-Ray GPU: https://www.jonasnoell.com/downloads/vray/gpu_vray.mp4
    FStorm: https://www.jonasnoell.com/downloads...gpu_fstorm.mp4

    If anyone has anyhints of how to reach faster speeds with V-Ray GPU I'm all ears
    Cheers!
    Check out my FREE V-Ray Tutorials

  • #2
    Nice technical comparison.
    Impressive work from Fstorm.

    I look forward to a response from Chaos.

    I'm worried to see more and more topics/tutorials on Corona and FStorm than on VRay.
    I get the impression that VRay is losing popularity.
    Maybe I'm wrong. I hope I'm wrong.​
    Windows 10 - RTX 3090 - AMD Ryzen 9 5950X - 128 Go RAM
    Sketchup PRO 23.0.419 + V-Ray 6.20.02
    3DS Max 2024.2.1 + V-Ray 6.20.06
    Chaos Vantage 2.3.0
    GeForce Studio Ready Driver Version 546.01​

    Comment


    • #3
      Simple scenes are often faster for dedicated GPU renderer.
      Try again with a real production scene. With gb of data, displacement, motionblur, dof, VDBs and fur.

      If thats not your typical stuff you render go with fStrom.
      https://linktr.ee/cg_oglu
      Ryzen 5950, Geforce 3060, 128GB ram

      Comment


      • #4
        Originally posted by oglu View Post
        Simple scenes are often faster for dedicated GPU renderer.
        Hi, I was under the impression that V-Ray GPU is by itself a "dedicated" GPU renderer. At least Chaos frequently emphasizes that both engines for CPU and GPU are completely separate.
        You are right though, so far I could only test it on a very simple scene. The more complex a scene gets the more difficult it would be to compare as there are many more variables and ressources that would have to be matched in order to compare everything fairly. If you have any example you could point me to I would be happy to check that out.
        Here I am merely trying to compare the pure render speed, nothing else. As stated earlier V-Ray offers much more compelling features and a more production proven trackrecord (at least for CPU).
        Still I think it is very remarkable how much faster FStorm performs in this example scene. For me personally the difference is big enough that in terms of GPU rendering I would rather experiment with FStorm and see how much I can implement this in my day to day work.

        Attached you can find the same scene rendered with FStorm, Redshift and V-Ray GPU. Same hardware, same ressouces, same resolution. I tried to also have same trace depths for each render engine, though somehow the refractive materials looks slightly different. I could experiemnt more to get a closer look but it doesn't seem to change much in terms of performance. I tested with different types of materials, using same or (hopefully) very similar settings across the engines.
        The result is always consistent: FStorm outperforms the other engines by a mile. Redshift comes in at second with much finer and less contrasty noise. V-Ray always has the highest amount of noise at the end. Even when using super simple materials such as just a grey dummy shader. Rendertime was always roughly 20 seconds and from my tests it would take V-Ray a multitude of time to reach a comparable result in terms of image quality.

        You can watch a video here:
        https://www.jonasnoell.com/downloads...shift_vray.mp4

        If anyone knows or has any ideas how I could get a similar performance in V-Ray please let me know
        Cheers
        Attached Files
        Last edited by JonasNöll; 27-01-2024, 04:07 AM.
        Check out my FREE V-Ray Tutorials

        Comment


        • #5
          As Oglu already noted you should try with a complete scene, results from simple scene or benchmarks can be misleading. Don’t know about FStorm, but at least comparing RS and Vray GPU I’ve seen similar results in the past in terms of render speed. Beside this always test with you typical scene, this will not only give you a more accurate speed comparison, but it will also give you the chance to test if the feature set is up to your needs.
          3D Scenes, Shaders and Courses for V-ray and Corona
          NEW V-Ray 5 Metal Shader Bundle (C4D/Max): https://www.3dtutorialandbeyond.com/...ders-cinema4d/
          www.3dtutorialandbeyond.com
          @3drenderandbeyond on social media @3DRnB Twitter

          Comment


          • #6
            JonasNöll one thing I would recommend making sure is set more appropriately for your GPU configuration is the "IPR rays per pixel". This doc might help ya: https://support.chaos.com/hc/en-us/a...g-in-V-Ray-GPU

            In my experience it can speed simple scenes up quite dramatically

            Comment


            • #7
              Originally posted by Nejc Kilar View Post
              JonasNöll one thing I would recommend making sure is set more appropriately for your GPU configuration is the "IPR rays per pixel". This doc might help ya: https://support.chaos.com/hc/en-us/a...g-in-V-Ray-GPU

              In my experience it can speed simple scenes up quite dramatically

              Hi, thanks for this hint! Actually from first testing that does seem to improve the performance indeed quite drastically! I will do some more testing tomorrow to see how that compares then.
              How does that work if you want to submit the scene to a renderfarm with very different GPU configuration? As this is saved in the rendersettings of the scene it would mean that some GPU render with optimal settings and some are forced to use settings intended for different hardware?
              Check out my FREE V-Ray Tutorials

              Comment


              • #8
                Hi, so after tweaking the "IPR rays per pixel" to the recommended settings from the mentioned docs it improves things a lot. Attached you can find the updated test where in this (simple) example V-Ray now clearly leaves Redshift in the dust (though while still not reaching the speed like FStorm). Redshift doesn't really offer any options in the settings that could tweak this or otherwise influence the IPR performance apart from the Ray depth limits, otherwise I would also like to have tried that.

                https://www.jonasnoell.com/downloads..._vray_v002.mp4
                Check out my FREE V-Ray Tutorials

                Comment


                • #9
                  Originally posted by JonasNöll View Post

                  Hi, thanks for this hint! Actually from first testing that does seem to improve the performance indeed quite drastically! I will do some more testing tomorrow to see how that compares then.
                  How does that work if you want to submit the scene to a renderfarm with very different GPU configuration? As this is saved in the rendersettings of the scene it would mean that some GPU render with optimal settings and some are forced to use settings intended for different hardware?
                  Awesooome, glad to hear that helped

                  The IPR rays per pixels affect the IPR rendering only. For "final frame" rendering you've got the sans IPR "rays per pixel" parameter and so when you send things to a render farm the IPR settings won't really have an effect.

                  From my understanding what the rays per pixel parameter does is with higher values keeps the GPU busier and thus it improves its utilization. You'll probably notice that with lower values the GPU itself will report much lower utilization (GPU-Z / Task Manager / whatever you like to use) and thus the rendering is actually slower than it needs to be. With higher values that then doesn't really negatively affect "slower" GPUs, they'll just be 100% busy.

                  That said, what higher values will change in the IPR for you is how "fluid" the feedback will be. With higher values each "pass" will look much cleaner (noise wise) than with lower values but then because each "pass" now takes longer the experience will be less "fluid" (think of it like sampling the image more in same amount of time - broadly speaking). So ideally when you want quick updates in the IPR you are near 100% GPU utilization but are also not overdoing it to an extent where you move the camera and then wait 2 seconds for the image to update.

                  The values on that doc I linked too work pretty well for me personally (similar setup as yours, 4090 x 2)

                  Speaking of render farms, if you're using V-Ray DR then it might be beneficial to render using the bucket sampler type - even on GPUs. Currently more often than not DR just works faster with the bucket sampler type. And the same holds true if you're rendering locally on three or more GPUs (https://docs.chaos.com/display/VMAX/GPU+Render+Settings).

                  Also, huge fan of your tutorials, keep up the awesome work man!!

                  Comment

                  Working...
                  X