Page 3 of 12 FirstFirst 12345 ... LastLast
Results 21 to 30 of 115

Thread: Maya 2015 and mental ray GI GPU prototype

  1. #21

    Default

    Such claims have no value unless you share scene for others to compare. Anyone can slap here a random picture with rendertime sucked out of their thumb. And viewing fullres version of that image shows severe blurriness. I hope that's just overshot gaussian filter and not postprocess denoising.

  2. #22
    Join Date
    Dec 2005
    Location
    Wherever The Computer Says
    Posts
    2,853

    Default

    Or you could just render the test scene he's already provided....

    I'll be more than happy to talk about the other renderer after SIGGRAPH....
    Last edited by Remydrh; July 20th, 2014 at 10:26.
    "Don't argue with an idiot, they will drag you down to their level and beat you over the head with experience."

  3. #23

    Default

    Seriously? You want to compare your VFX production ready renderer on a cornellbox scene? Three primitives with diffuse gray material in a diffuse box? I can sure do that, but the defeat will be that more devastating for mental ray.

  4. #24

    Default

    BTW, i do not even think there's more than 1 bounce of GI in your scene. It looks a bit fake when it comes to indirect diffuse interaction between surfaces. What i see is scene heavily lit by direct lighting. Scene designed so that it shows as little indirect illumination as possible.

    Look at here for example
    GI_BS.jpg
    The area under window is obviously strongly illuminated, yet there is no bounce spill on the wall perpendicular to it, even when the wall is connected to the illuminated area. As if there was no bounce light at all.

  5. #25
    Join Date
    Dec 2005
    Location
    Wherever The Computer Says
    Posts
    2,853

    Default

    If there was no bounce light it would be black, not grey/blue from exterior spill light. You're used to lighting scene with lots of indirect bounces for architectural images. Also, you can plainly see the indirect buffers from the EXR I posted publicly in the first section of posts.

    Maybe you missed that....

    We use 1 or 2 because we're usually only concerned with the first diffuse bounce in VFX. This is the typical settings in mental ray, Arnold, and Renderman. Adding additional bounces is less painful anyway but unnecessary for us since we don't care.

    So you're just not used to seeing that I suppose. In this case your observations about my settings are wrong based on your personal experience.
    Last edited by Remydrh; July 21st, 2014 at 02:05.
    "Don't argue with an idiot, they will drag you down to their level and beat you over the head with experience."

  6. #26
    Join Date
    Dec 2005
    Location
    Wherever The Computer Says
    Posts
    2,853

    Default

    Funny thing is there is indeed a huge artifact from an error in this image. It has nothing to do with the new GI and is a result of a Maya shader.

    No one seems to be able to notice it despite all the other actual non-flaws. (And it's pretty obvious to me anyway, I even verified it testing the scene in the other renderers)

    Makes me laugh no one has said that's not right. Although now that I say it....

    I will post the fixed version tomorrow. I just had to sanitize a couple Maya parts from the scene that haven't been updated in 2015 non-service packs.
    "Don't argue with an idiot, they will drag you down to their level and beat you over the head with experience."

  7. #27

    Default

    Ok, since no one has provided some better scene, i did at least the cornellbox test.

    CPU used is Core i7 3930k, which is priced about same as GTX670 GPU. You should also keep in mind that i am rendering ONLY on CPU, so to do fair comparison, you need to take in account that Actindia used both CPU and GPU when rendering.

    Another two conditions you should keep in mind:

    1, You can clearly see the strong step between noisy and completely clean area (i mean extremely clean, no noise even when hard contrast adjustment is done) that implies he used interpolated final gather, not brute force... because if brute force was used, then adaptivity would make sure there is at leas some noise (invisible to naked eye) to be efficient and not waste samples under contrast threshold. But there is none, so interpolated FG was used. This is very problematic, because Arnold is a brute force path tracer, which makes this comparison very unfair.

    So in my test, I used Irradiance Cache to make the comparison with mental ray fair.

    2, You can clearly see jagged edges on Actindia's render, which implies insufficient antialiasing. Quality that is unacceptable to turn over to client, so final quality would take even longer, than the times he wrote.

    So here is my result with Vray 3 on 3930k:

    10 seconds
    Vray.jpg

    Now i would love if someone shared some more complex scene, so i can embarrass Mental Ray even more.

    I am also attaching 3ds Max 2014 + Vray 3 scene for anyone to verify my results.

    VrayCornellBox.zip

  8. #28

    Default

    Oh, and as i can see already someone trying to discredit my results saying that my render lacks contact shadows, I can assure you that it is only due to the increased albedo (diffuse reflectance) of the gray material, which is higher than what Actindia used. Here is render with darker gray color, where indirect shadows are more pronounced. Decreasing diffuse reflectance increased overall scene contrast, so adaptive sampler in Vray had more work. As a result, rendertime increased to 13 seconds, which is IMHO still quite a good result.

    Vray_Darker.jpg

  9. #29
    Join Date
    Apr 2009
    Posts
    511

    Default

    you used interpolation gi from vray to compare with a brute force gi solution that newgi is?
    besides the fact that you had to lower the color intensity to get contact shadows.

  10. #30
    Join Date
    Feb 2013
    Location
    Hannover, Germany
    Posts
    115

    Default

    Quote Originally Posted by rawalanche View Post
    1, You can clearly see the strong step between noisy and completely clean area (i mean extremely clean, no noise even when hard contrast adjustment is done) that implies he used interpolated final gather, not brute force... because if brute force was used, then adaptivity would make sure there is at leas some noise (invisible to naked eye) to be efficient and not waste samples under contrast threshold. But there is none, so interpolated FG was used. This is very problematic, because Arnold is a brute force path tracer, which makes this comparison very unfair.
    It's NOT interpolated final gather.
    FG was turned OFF.
    As I wrote it's pure GI GPU. So both, arnold as well as mental ray are brute force in terms of GI.
    But I noticed that low GI GPU itsn't really producing noise. Instead t's looking cloudy... Maybe someone can explain how the new GI is working...

    10 sec for your vray test is nice. But to make it fair you should use some brute force algorithm.

    I'm currently testing with a much bigger scene... I will provide some footage and the scene as well as soon as possible.
    Last edited by Actinidia; July 21st, 2014 at 12:45.

Posting Permissions

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