I’ve got a bit of downtime which I’m using to learn more about the vray sampler.
Three things that have surprised me,
1: Using the DMC sampler, 1/4 gives cleaner results than 1/24 . I would have thought the higher upper limit would result in a better result. I’m using QMC GI first bounce and LC second.
2: Adjusting the DMC_NT (Noise Threshold) affects the light distribution of the LC. I had always assumed the light cache parameters were self contained (ie not affected by the DMC sampler)…
3: Using just QMC for first bounce(adaptive 0.95, NT 0.01, clr thresh 0.001; aDMC 1/4 gives a decent result… 1/8 darker image and more noise… 1/18 back to the lighter result of 1/4 and less noise. Why does the result decrease and then increase in quality as I increase the max samples? I would have expected a more linear progression from high to low levels of noise.
I know I’m exposing my ignorance on this thread, but what the hell! It just goes to show how much I still need to learn!
This is true but in a sense it’s a little frustrating in that you can never develop a solid theory about how to optimise your renders - in this case it’s probably something like if you use less aa, the dmc sampler uses higher samples before passing its results to your aa which are nice and clean - with aa 1 / 18 it thinks it needs less dmc samples to achieve the same level of cleaness so the gi solution ends up being worse before it passes on a result to the aa section - in this regards the unified architecture is a little confusing. Again I could be getting the theory totally wrong but I agree with you that it’s a tiny bit counter intuitive at times. It’s great having the flexibility but in a lot of cases on short turn arounds I just have to hit render without the time to do optimisation - Vlado mentioned something about some kind of sampling readout to aid in this but not sure where it is with all the other stuff he has planned.
On a related note, is it right to say that using a minimum aDMC rate of 1 is BAD… I find it ends up often ignoring AA samples at the corners of the buckets. Is the only way to avoid this by using minimum aDMC rate of 2?
