Squeezing out more render time

[treddie]

I think I might have stumbled on an interesting Octane "feature". At least this SEEMS to be true from my initial tests. Perhaps someone from the Octane team can chime in, here.

Anyway, each Octane render is given a time limit within which to complete its render. All noise may not be cleared up within that time so it is often necessary to do another render (which has a different seed value) and combine the two later in Photoshop. But I noticed that when I added a specular material with dispersion that of course, the render time doubled. So did that extra time get spent on just the refraction through the glass? That can't be since the entire scene is in GPU memory and ALL of it is getting worked on, I assume. Or did it then mean that everything else not related to the specular material would render twice as slow? It does not appear so. The side benefit is that although the render time is doubled to handle the specular material, it means I get twice the amount of time spent on all the other stuff, too, which means a much cleaner first render, overall, except for maybe that specular material.

If this is all true, then it would be possible to add a hidden (not to be rendered) mesh which has some quality to it that demands more render time...Like a specular material with dispersion, SSS, etc. This would force the render time to increase and allow for much cleaner first renders. The "hidden" material could be outside the render frame, or be hidden behind, or inside some other mesh so that it is not visible in the render.

It does not mean that if you get twice as much time, you get twice as much noise cleanup...Noise cleanup occurs logarithmically, and that specular material is taking up a lot of clock cycles as well. But nevertheless, it appears you DO get extra "time on the clock".

My first experiment seems to bear this out...My non-specular surfaces are much cleaner, given that extra render time.

[FooZe]

Hi Treddie,

Sorry to say, but this doesn't make much sense to me
Noise is related to the number of samples per pixel (in general). If you let both the base case scene and the base case scene with the hidden material render out to the same samples per pixel, you should find the base case scene should reach the given samples per pixel first.

By adding elements to the scene you should never be able to speed up the render.

All renders don't run for a specific time, they run until they hit a specific number of samples per pixel. Increasing this will mean the render takes longer to complete and will be more noise free when it is complete.

It is true that some things will effect the amount of rays that are sent in a particular direction (therefore seemingly changing the amount of "time" spent on different parts of the scene).

Things like:
1) emitter sampling rates (emitters will be 'sampled' based on their relative sampling rates). Note this sampling is not the same as a pixel sampling - it is something else.
2) HDRI importance sampling - parts of the HDRI that contribute more than others will be sampled more.
3) PMC will spend more time on rays that contribute to the image in a significant way (like caustics).

Hope this helps...

Thanks
Chris.

[treddie]

No, not speeding up the render. Just forcing it to cook longer. I just overlapped two renders in Photoshop and toggled the one on top on and off. The amount of cleanness in the base case + specular material is definitely smoother than the one underneath. Not by much, mind you, but definitely to the point that you can unnamiguously see the difference.

The first scene was the base case. The second one actually had 13 extra parts, one specular. 12 of those parts are small and under a glass window *the specular material)...They have about zero interaction with the rest of the scene. So with all that overhead, I still see a noticeable improvement given the much longer render time.

I don't doubt what you are saying, so something else is up. Maybe it is a slight amount of extra light bouncing off the big blue top part. But all the parts in the base scene cannot really "see" that big blue part...It is probably 95% hidden by the white card underneath it.

[FooZe]

Ahh, I think i see what you are getting at.
After X many samples per pixel, scene 2 is cleaner than scene 1, even tho' it is more complex.

The problem is, if you increase the samples per pixel so that scene 1 renders for 51hrs as well, you will probably find it is a lot cleaner than scene 2 after 51hrs. So really you aren't gaining anything...

I think the effect you are seeing is the difference in how the rays are interacting with the scene. Rays might be bouncing to the completion of their max depth in the darker areas (rather than a lot of time spent on rays that just bounce off to the environment or some such). For noise to reduce, rays results must more similar than before (if that even makes sense!). So something about the bouncing must be more unified or something (just a theory). Generally this is not something you can intelligibly exploit (especially not over something like PMC's ability to 'intelligently' choose meaningful ray paths).

Thanks
Chris.

[treddie]

That makes sense.

So my Unified Theory of Octane Gravity (UTOG) is bunk. DAMN! I was hoping it would lead to a new energy source!

[FooZe]

Or even better, a new time source! Unfortunately breaking the laws of space-time isn't that easy

[treddie]

All I can say is....DAMN!

[FrankPooleFloating]

Well, you did have 51 hours to come up with this hypothesis... and twiddlin' thumbs is only good for a couple hours, tops... time for an upgrade Treddie.

[treddie]

Heheh...I am so dying for a Titan. Just one!