Search found 42 matches

by Zelcious
Thu Jul 18, 2013 11:57 pm
Forum: General Development
Topic: Optimization strategies
Replies: 12
Views: 10301

Re: Optimization strategies

Are you saying that many shorter chains is much worse than fewer longer chains? On what scenes? I find it a bit hard to believe that having a larger number of chains is always worse. A tiny lightsource in combination with SDS path will certainly do. Only small mutations will be efficient and will r...
by Zelcious
Thu Jul 18, 2013 9:51 pm
Forum: General Development
Topic: Optimization strategies
Replies: 12
Views: 10301

Re: Optimization strategies

Perhaps the solution would be to let the cpu do the mlt and the gpu just looking for seed paths. For really really hard to sample scenes like some of the ones I experiment with it could perhaps be a winner.
by Zelcious
Thu Jul 18, 2013 9:16 pm
Forum: General Development
Topic: Optimization strategies
Replies: 12
Views: 10301

Re: Optimization strategies

This is a serious problem with MLT and GPUs: my implementation of MLT on GPU (with 65K chains) looks pathetic when compared to same MLT code on CPU (with 12 chains, i.e. 12 threads) . I hope multiple-try will remedy some of that. Plan to experiment with some higher order ensamble mcmc as well. I ha...
by Zelcious
Wed Jul 17, 2013 8:03 pm
Forum: General Development
Topic: Octane PMC
Replies: 0
Views: 2860

Octane PMC

I'm a bit curious about octane pmc implementation. Anyone care to speculate on how they use pmc? It's probably going to be for per pixel noise reduction and perhaps for selecting light sources etc but my guess is that they don't use something similar to pmc-er. What do you think? Could they use some...
by Zelcious
Wed Jul 17, 2013 7:33 pm
Forum: General Development
Topic: Optimization strategies
Replies: 12
Views: 10301

Re: Optimization strategies

I currently use 16k items in parallell. I take a slight hit with a gtx 780 but it's worth it since I use a mlt like algorithm and it converges faster with fewer chains. I'm heavily compute bound, use no resorting and split too long traversals to reduce incoherence.
by Zelcious
Sat Jul 13, 2013 2:47 pm
Forum: Links & Papers
Topic: Haswell benchmarks show up
Replies: 20
Views: 18335

Re: Haswell benchmarks show up

I got my new asus gtx 780 with directcu ii cooler yesterday and with a heavy overclock it's twice as fast as my gtx 480 with my own renderer. So far no tweaking have been able to remedy that. Quite a bit of disappointment actually.
by Zelcious
Thu May 30, 2013 10:09 am
Forum: General Development
Topic: Mutation questions
Replies: 0
Views: 2508

Mutation questions

Does anyone have any objections against: * Using the contribution of the full path and all subpaths containing the eye, each scaled by the density change up to that point when calculating acceptance probability. * A pertubation where you have the same number of diffuse vertices but allow the number ...
by Zelcious
Thu May 23, 2013 10:55 pm
Forum: Visuals, Tools, Demos & Sources
Topic: New renderer, WIP
Replies: 2
Views: 7263

New renderer, WIP

Started working on a new renderer. Very early WIP shot. Lots and lots of things to implement before it's correct. Like proper mutation probabilities, correct spectrum distribution instead of a made up one. Better intersection code and so on.... Also lot of optimizations and additional mutation strat...
by Zelcious
Thu May 16, 2013 8:10 am
Forum: Links & Papers
Topic: SIGGRAPH 2013 papers
Replies: 6
Views: 6909

Re: SIGGRAPH 2013 papers

Im mostly interested in the paper: Gradient-Domain Metropolis Light Transport. I hope it's good. But it's probably more about noise reduction than my initial thought about something similar to manifold exploration that I have been thinking about. (you should be able to do a numerical search based on...
by Zelcious
Tue Feb 19, 2013 7:16 pm
Forum: Links & Papers
Topic: Implicit and Dynamic Trees for High Performance Rendering
Replies: 7
Views: 6862

Re: Implicit and Dynamic Trees for High Performance Renderin

Is there somewhere i can get a free copy? Only found pay walls so far but im also searching from my mobile...