Subdiv optimisation for RT

[3LP]

Hi guys,
I'm using RT for final preview as well as final renders now.
It's working pretty well but I have often noise that takes ages to clean on reflective wall for example.
I was wondering if there was any way I was able to force somehow subdivs on few materials?
I know the subdivs are discarded by default from the way RT is working, but even is there still a way to get those in? or MSR?

I often have my render looking great after 5-10 minutes but I have to leave it rendering to clean those few materials that are noise and even after 1h it's often still not clean enough.

I haven't tried on RT CPU but I'm using RT GPU. If it was for the cpu, I could use the normal production render.

Stan

[vlado]

Can you get me an example scene to vlado@chaosgroup.com that is giving you trouble? There are a number of things that we can do to improve this, and I want to check how they work with your scene.

Bets regards,
Vlado

[super gnu]

i have the same issue.. scene looks passably good after a couple of minutes, nice after another couple but to get it really clean (i.e good enough for animation) takes 10x longer, to the point where maybe standard vray would be comparable.
i dont even need to send you a test scene. the standard rt gpu benchmark works perfectly to illustrate this.

ive tried various combinations of rays per pixel and noise threshold limits... in fact ive noticed whenever i have a noise threshold other than 0, i seem to get more hard to shift noise (in the really difficult areas, not the cleaner areas that are being stopped refining by the noise threshold)

however if i understand correctly, a noise threshold of 0 means it continues sampling even the clean areas, instead of focusing on the more noisy ones, so is inefficient.

[vlado]

Can you send me such a scene?

Best regards,
Vlado

[super gnu]

well i can, but its just your rt gpu benchmark scene.. switching it over to the rt production renderer, setting res a bit higher (i was using 1280 x 720) and trying to get a clean result using various combinations of noise threshold and paths per pixel. (i dont think it matters, but i was using LC/LC for lighting, as per my nice imap in rt thread)

i can get reasonable result in about 4-5 mins there, but to get it clean(ish) takes 15 - 20 mins, and i didnt bother trying to get a noise-free result, as i got bored, but its over 30 mins.

[vlado]

Ah ok. Yeah, I know why that is - we are working to improve it.

Best regards,
Vlado

[super gnu]

cool. yes its odd, i would expect that as the render progresses, noisier areas would clear up increasingly fast, as the rest of the image arrives at the noise threshold and stops being updated, leaving the gpu focusing its energies on progressively smaller areas.. maybe this is happening, but it doesnt appear to be to any extent, the difficult areas are very stubborn to shift, and past a certain point dont seem to get any cleaner, or do so increasingly slowly, despite obviously still being above the noise threshold.


this could of course be an optical illusion, maybe the improvement is just much more noticeable visually at the beginning.

[vlado]

Ok, that scene is very old, it's possible that it's not set up with correct color mapping. For other scenes, it looks like the noise threshold works as expected.

Best regards,
Vlado

[super gnu]

what is the correct setup for colour mapping in this case? i can check.


by way of comparison, having been very excited by the speed of RT when using precomputed GI (it -is- fast) i tried rendering the same scene in the standard renderer on a 6 core i7.

i got a pretty perfectly clean frame in about 9 minutes. so.. that would be faster than rt gpu on a 670 4gb (scene was still not clean after 15- 20 mins) (and extrapolating approximately, also similar or faster than a 980)

im not sure wether to be impressed with vray 3 or dissapointed with RT gpu in this case!

probably i need to try a different scene.

[vlado]

The default settings - 3ds Max gamma 2.2 enabled, V-Ray color mapping gamma 2.2 (but without the gamma applied).

Best regards,
Vlado