Since I am using OptiX to build a pathtracer (based on Detlef’s OptiX Advanced Introduction Samples) I tried a lot of methods to create good noise.
I read somewhere, that Blue Noise would converge faster, but that White Noise would result in better quality when long enough iterated. Can someone confirm that?
(My created Blue Noise obviously yet is not good enough to show any different effects than white noise.)
For the beginning, I would try XOR scrambled Sobol. That should give you a good convergence and is easy to implement. I believe curand already includes an implementation.