Vray - also a fast one button render engine?

[Micha_cg]

Re: Vray - also a fast one button render engine?

Hi Gijs,

interesting for me is:
- The adaptive calculation could give the chance to keep anything so simple as possible. I would wish more development research by Vlado here.
- Adapative calculation can be fast, maybe some engine improvements are necessary. I suppose, if the full adaptive workflow show problems, than they can be solved.
- I find two different kind of noise: low frequency IM noise and high frequecny fine QMC noise. This two kind of noise should be independent controlled. It cause problems, that this two noises are dependent.

EDIT
I did some tests more. It seems to be, that the IM create smoother results at large surface without lighting details in 85% adaptive mode. In 100% mode the large surface show light splotches, the IM take to less samples. I think Vlado could optimize the IM for full adaptive mode.

[Travis Serio_cg]

Re: Vray - also a fast one button render engine?

Interesting, good find Micha. I'll try it out as well and see if i get same. I am the splotch master ;D

[Gijs]

Re: Vray - also a fast one button render engine?

I applied the settings from the test scene to a different (more complex) scene, but even in gray scale rendering, I can't really say that this method is universal. Actually the result was fast but splotchy, especially near small edges and large surfaces turned out quite splotchy too.
We better wait until next SR which incorporates detail enhancement, this has become my favorite choice in 3dsmax.
In my opinion you can spend many hours trying to find the optimum IM settings for a certain scene, but since it is an approximate method and highly dependent on your environment, I don't think it is possible to find a universal setting.
The biggest problem with IM is that you get noisy results at higher settings on large surfaces. So it's either a choice of lower settings and a smooth, not so detailed GI solution, or wait for detail enhancement I vote for the last.

[Micha_cg]

Re: Vray - also a fast one button render engine?

Gijs, I saw this effect at large surfaces too. Maybe this can be optimized from the IM code.
An other kind of universal setup could be, that all fine noise subdivs are working in full adaptive mode and the IM independent works in 0.85 adaptive mode.
I'm very curious for the detail enhancment.

[dalomar_cg]

Re: Vray - also a fast one button render engine?

I have been messing around with this again for the last few days, and I have a few observations/comments or whatever. First off, the viewport update is a real necessity for this...without it, the only way this is really useful is for a final render because there's no feedback. Secondly, I am wondering if there is any advantage (speed and/or quality) to using LC over QMC for secondary bounces. I had personally thought it would be simpler just to leave both passes on QMC (so that the render could literally be killed at any moment) and adjust the amount of bounces for qmc. I will go through and try a few tests and see if there is any advantage.

Thirdly, it would be really great to have the indirect calculations "overlayed" over the direct light calculations. In other words, calculate all of the direct illumination first (simple and fast) then refine all of the indirect illumination based on that. This would prevent scenes with a large amount of direct light (or simply areas of a scene) to not rely on the qmc pass to determine the large amount of their illumination.

Also I ran into an intersting speed bump that may or may not actually be a speed bump. I started thinking that if I had a multi-core computer, would it be possible to render multiple images at the same time (ie if I had a dual core, rather that set the bucket size to half the image size and have both cores work on the same image. I can set the bucket to the full size of the first image and have another core free for another render), so that I didn't have to kill one just to start the other. However, even with batch render disabled, it is not possible to have two renders happening within the same rhino task. This means that you would have to open up another rhino task in order to render another image. At first I thought this was an issue, but in lieu of 64 bit windows the having multiple rhino tasks will increase the total memory headroom because each task can access up to 1.6 or 2 gb (or whatever it is I can't remember). Anyway just a interesting discovery.

Lastly, as far as speed goes, I almost never use 100 subdiv for aqmc image sampling. 24 works fine for me, and I would argue that there aren't many cases where you'd really see the difference. Oh and also 1000px is the maximum size for buckets, which limits the possibilites of using this method for higher res images

[Micha_cg]

Re: Vray - also a fast one button render engine?

Yes, the refresh of the framebuffer durung the calcualtion must be. The current test is good for experiments only. Interesting idea to use full QMC. I'm curious how it works.

I like to use the LC, because it is best physical correct and support biased shortcuts like "use for glossy reflection" and prefilter. Also I like the fast full preview. But with progressiv full QMC this would be the same fast preview.

The 100 subdivs are good for special difficult calculations, like high intensity reflections (-> sun problem). But, than the render time is quite long at this parts of the image. A lower subdiv value is a good render time limit.

Thanks for taking time to experiment.