Announcement

Collapse
No announcement yet.

Bizarre render performance

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

  • Bizarre render performance

    I have an interior scene with cached IR and LC.
    At quality settings aDMC 1,20 Subdiv Mult 0.5 I get a nice render.
    If I drop the quality to aDMC 1,3 Subdiv Mult 0.4 I get a lower quality render, to be expected.
    HOWEVER, the lower quality settings are about 30% SLOWER to render ??!!!??
    Can anyone explain why ?
    Thanks !

  • #2
    the maximum dmc subdivs will affect how the subdivs in your scene are actually evaluated.. raising the max dmc subdiv value will, in effect, reduce subdiv values in your scene elsewhere and rely more on eye ray samples instead. in your situation, you most likely have a very high subdiv value somewhere that is weighted more heavily when the max dmc subdivs are lowered. this is counter-intuitive, but it's what makes the unified sampling so powerful in complex situations.

    for example:
    at dmc 1,1 a glossy reflection subdiv value of 256 will fire 256 subdvis
    at dmc 1,2 a glossy reflection subdiv value of 256 will fire 128 subdivs (to start).

    as you raise the max dmc value, you are basically reducing the subdivs in your scene and allowing the sampler to decide whether more subdivs are needed (based on the noise threshold). for this reason, it's not uncommon to actually get *cleaner* results when the max dmc is lower--it all depends on what is creating noise in the scene.

    Comment


    • #3
      Thank you so much ! Very informative !
      Two questions,
      1) What are 'eye ray' samples ?
      2) What would you consider a high subdiv value for glossy reflections ? I tend to use 8,16,24 & 32 (32 only if absolutely necessary)
      Thanks again !

      Comment


      • #4
        1) by "eye ray" i mean an initial sample that goes from the "eye" (camera) into the scene. after it hits the scene, vray sends additional "subdiv" rays for fuzzy values (like glossy reflections, area shadows, motion blur, depth of field) based on settings\thresholds.
        2) if you are using adaptive dmc sampling, your dmc settings (max) will determine what is a "high" glossy reflection. (it also depends on how glossy your surface is)

        there are many different approaches to a clean render, but in general i use "medium" dmc settings (1,16-32) and i tend to leave scene subdivs low (--only adding subdivs if i see noise. if i see noise (in a glossy reflection, for example) at 1x16, i know i'll need at least > 16 subdivs to get more subdivs to fire. if i want 2 or 3 (instead of 1) per "eye" sample i'm going to need 16 x 2 or 16 x 3 (32,4. this is a slight simplification of what's actually going on, but it helps predict what the sampler will do. in general for the glossiest reflection, i will usually not go above 64 subdivs; but again, this is only relevant when using a specific maximum dmc value.

        Comment


        • #5
          Originally posted by gimiky View Post
          there are many different approaches to a clean render, but in general i use "medium" dmc settings (1,16-32) and i tend to leave scene subdivs low (--only adding subdivs if i see noise. if i see noise (in a glossy reflection, for example) at 1x16, i know i'll need at least > 16 subdivs to get more subdivs to fire. if i want 2 or 3 (instead of 1) per "eye" sample i'm going to need 16 x 2 or 16 x 3 (32,4. this is a slight simplification of what's actually going on, but it helps predict what the sampler will do. in general for the glossiest reflection, i will usually not go above 64 subdivs; but again, this is only relevant when using a specific maximum dmc value.
          Thank you for sharing your experience!

          I am trying to learn more about fine-tuning the Adaptive DMC setting for a scene. If I understand correctly, the way you suggest to tune the setting is by setting scene subdivs (like glossy subdivs of materials and subdivs of lights) at default 8, and then cranking up max adaptive subdivs from a "medium" value like 16 to 32 or 48 when you see noise. Can this approach be applied to noise seen in other elements like Raw Light and Raw GI?

          How would you decide if one should crank up subdivs of individual materials (for glossy reflection) and light, instead of the max subdivs of adaptive DMC?

          For me, it's kinda tricky and confusing on how to find the "balance" subdivs settings among the three: 1) materials and lights, 2) max subdivs of adaptive DMC, and 3) global subdivs mult under the DMC sampler. The "balance" above refers to getting desired clean render efficiently without sending too much samples that increases the render time. It will be wonderful if you could shed some more light in this regard.

          cheers.
          always curious...

          Comment

          Working...
          X