Announcement

Collapse
No announcement yet.

Performance Comparisons

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

  • Performance Comparisons

    Alrighty! I spend a good day experimenting with RT-GPU and was overall pleased with the results. There was though an interesting test I did which gave me some experiences I'd like to share for the eternal bug-hunt.

    First my critical system's specs:

    DELL Precision Workstation T7500
    Dual Xeon E5530 2.4ghz Quad-Core, Hyper-Threaded 8mb L3 cache 1mb L2
    24GB DDR3 Memory
    Dual NVIDIA Quadro FX 4800 cards with 1.5GB memory each
    4 x 420gb 15,000rpm SAS drives in RAID array
    Windows 7 Ultimate - 64bit
    3ds MAX 2011 - 64bit

    Anyhow, here is the scene rendered with V-Ray Production. (this is the kitchen from my new apartment)



    Nice, clean and quick.

    Here is the same scene pushed through V-Ray RT CPU:



    First of all, the intensity and warmth shifted noticeably but it did take less time to get a satisfying image for making adjustments.

    Here is the same scene using V-Ray RT GPU:



    First of all, as the stats indicate, I was getting over six times the "Kpath/s" performance, but even still, I feel the image was far less refined than the CPU render.

    The CPU render boasted 56 samples/pixel reached and managed a cleaner image than the 384 samples/pixel reached.

    One thing of note, the GPU render was so under-exposed compared to the CPU, I had to change the shutter speed and "ISO" of the camera to compensate.

    You can download and test the scene here: On my Website

    Still, with my other experiments, I am very hopeful for the excellence that is V-Ray 2.0 both CPU and GPU!

    Thanks,
    EJDII

  • #2
    U can use both CPU and GPU right now too, Just need to do quick tweak... Ill get a a link in a minute... About render speed it looks like ur GFX its slower than CPU... u sure they work right and u have newest drivers?

    Last post page 2@Vlado
    http://www.chaosgroup.com/forums/vbu...049#post436049
    CGI - Freelancer - Available for work

    www.dariuszmakowski.com - come and look

    Comment


    • #3
      Hehe, nothing beats an optimized biased render Thanks a lot of the scene, will be very useful for debugging - I get a page not found error though.

      Quadro FX 4800 is somewhat on the slowish side though; newer cards are faster.

      Also keep in mind that the GPU doesn't support light portals yet (not sure if you used one or not).

      Best regards,
      Vlado
      Last edited by vlado; 09-09-2010, 03:13 PM.
      I only act like I know everything, Rogers.

      Comment


      • #4
        DADAL: I actually tried that mixed mode and it made no difference when I ran it. Perhaps I did the environment variable wrong.

        VLADO: I apologize I had the link wrong. It should be fixed now. And yes, I used VRaySky for the environment so that makes sense why I had to compensate for the lack of light.

        Yeah, the FX 4800 uses the GT200 family GPU so I don't expect much in the way of amazing improvement until Fermi and beyond. Maybe I'll swap out the cards for a pair of GTX 480s... It would still cost less than one Quadro 5000. (and more CUDA cores though less memory...)

        Has your experience been that after 1.5gb of memory, the more CUDA cores available on a GPU, the better the performance? Or would I be better suited to get a card with more memory and less CUDA cores?

        Comment


        • #5
          Just for fun tested it on a pair of 8800GTX

          Dual Opteron 2220SE, 16GB.

          Resolution set to 853x480px.




          Rendering was stopped after 2min


          Rendering was stopped after 14min. sorry about that.

          Conclusion: I need new graphic cards

          Comment


          • #6
            I haven't had a chance to test the said scene, but my quadro 1800 at work is so so so slow compared to my gtx 285 at home.
            Boss aintgonna believe me when I tell them I need a 'crappier card' to be more effecient

            Comment


            • #7
              So some notes:

              = The production renderer has no secondary GI engine, while the RT has 3 GI bounces (not sure if we should attempt to correct this);
              = The "No decay" option for the sphere light is not supported on the GPU right now (most non-physical stuff isn't);
              = Primary engine had multiplier 2.0;
              = There was no sun in the scene, only VRaySky (this seems to cause some differences between V-Ray RT and the production renderer; have to look into it).

              I'm rendering now with these corrected; will post the images in a moment.

              Best regards,
              Vlado
              Last edited by vlado; 10-09-2010, 04:33 AM.
              I only act like I know everything, Rogers.

              Comment

              Working...
              X