Page 1 of 2 12 LastLast
Results 1 to 10 of 11

Thread: Gpu rendering

  1. #1
    Join Date
    Dec 2016
    Posts
    2

    Default Gpu rendering

    HI,

    I was trying to test the new functionality of mental ray 3.14 GI next using the GPU to render.

    Even if the GPU seams to be using 100% of it's processing power, I don't see any improvement on my render times ( Final Gather renders even faster ).

    There is not that much information about this online.

    I am wondering if I need to configure anything before rendering or if my graphic card is compatible with this new system.

    I have to nvidia 1080's connected through SLI.

    I hope anyone can help with this issue.

    Thanks

    Antonio

  2. #2
    Join Date
    May 2011
    Location
    San Francisco Bay Area
    Posts
    374

    Default

    Hi Antonio,

    You never need SLI enabled to use multiple GPUs with a CUDA renderer - any CUDA renderer.
    We always advise disabling SLI when using multiple cards for compute - even if the two GPUs are on the same card.
    So please start with that, as SLI could be curtailing your GI-Next rendering speed considerably.

    Also, when you compare to Final Gather are you using FG Force?

    - Phil

  3. #3
    Join Date
    Dec 2016
    Posts
    2

    Default

    Hi Phil,

    First of all thanks for your replay.

    Sorry for the late replay but I did not had the chance to do more test until know.

    For the test I did today I disconnected the SLI function on the card.

    It is a very simple scene with two MILA material both with reflection ( one using weights at 100% and the other using fresnel ), also motion blur is on in both cases and there is only one directional light and a IBL with a 6k image.

    I am attaching 2 images where you can see the render times and the settings I was using for Final Gather and the GInext ( Final Gather Force takes even longer ).

    You can also see the cards I am using the number of cudas and the speed, as well as the CPU threats that I have ( I even used MSI after burner tool to make sure I was using the GPUs to their full potential and they where ).

    Here are the images I hope you can help me with this issue. If you think the rendering log will help let me know and I will post both.

    Thanks

    Antonio

    finalGather.jpgginext.jpg

  4. #4
    Join Date
    Dec 2004
    Location
    Marina Del Rey, California
    Posts
    4,143

    Default

    Hi, since GI Next and FG force have similar quality from brute force tracing techniques, that is why we were asking you to compare those.

    Finalgather is interpolating, so is a different quality technique that generally takes more tuning of the parameters. However, we do try to set a few (eg, Accuracy, Point Density) through the quality slider.

    With GI Next, please turn on the render report with verbosity level of "Progress", and copy and paste the section between GI sampling and regular tile rendering.

    Also, what is the time difference when turning on and off the GPUs for GI Next? You may notice more difference with higher GI Next quality, and deeper diffuse trace depth.
    Last edited by bart; February 13th, 2017 at 04:52.
    Barton Gawboy

  5. #5
    Join Date
    Nov 2016
    Posts
    6

    Default

    Hi all im linking on this discussion for the GPU renderings.

    i need to know that, i have an Quadro M4000 GPU, and during renders with GINext i have checked the GPU usage on the card panel
    the mid usage is around 50% with some peaks, but i whas expect to see the gpu utilization graph standing almost on 100% all the time like
    the CPU one in CP , during normal renders. Is this normal? and if not depends on some of my settings?

    Render_perf.JPGthank you in advance for your reply.
    Last edited by Nicolò; February 22nd, 2017 at 12:37.

  6. #6
    Join Date
    Feb 2017
    Posts
    1

    Default

    I have 2 quadros 5000, 92 processors and 192gb of ram to do architectural rendering an animation. At this point this is getting short for me, so next step is to by another server with 80 or more fast processors, but is it better to have one of those quadros connected to a tesla k80 for a maximus performance. Wich one will be faster? A bucket rendering with 250 xeon processors or a quadro k5000 plus a tesla k80?

  7. #7
    Join Date
    Dec 2004
    Location
    Marina Del Rey, California
    Posts
    4,143

    Default

    Perhaps this should be moved to a separate thread to address the tradeoffs.

    This depends on what you're doing with the scenes.

    Are they GI heavy scenes?
    What resolution?

    Is this aimed for getting a single frame rendered quickly? Where you might consider satellite?

    Or is this just about finishing an animated shot quickly?

    Are your 92 procs on a single machine? Physical or logical (hyperthreadaed) core count?

    Not quite sure what you mean by a Tesla K80 connected to Quadros. Do you mean adding a Tesla K80 to the machine that has 2 Quadros, or considering spreading the GPUs to different machines?
    Barton Gawboy

  8. #8
    Join Date
    Nov 2016
    Posts
    6

    Default

    Any advice for me?

  9. #9
    Join Date
    Dec 2004
    Location
    Marina Del Rey, California
    Posts
    4,143

    Default

    Yes, a few more scene details would help. It might be that GI Next is not used much in this scene.

    Are you lighting with HDRI and using IBL with emit light?
    Barton Gawboy

  10. #10
    Join Date
    Nov 2016
    Posts
    6

    Default

    Hi Bart thank you for the answer

    Yes HDRI IBL with light emitting, make some test with no light emitting too.
    Mia material X and some 4k light textures that i have made.
    Render time are really close to normal FG renders, just 10% less. I am missing somenthing?
    I was dreaming about render times became 1/4 - 1/5 compared to my modest 6 cores CPU, but actually the increasiing of speed
    is not much than 10% .

Tags for this Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •