Here are some bench results from, scenes are not converted or built for purpose of GPU rendering, there is some minor difference in appearance but that is due current limitations of engine.
RT is used as production render, it is BF/LC setup both for CPU and GPU. CPU is almost default with exception of 0.005 threshold, DR on 3 machines(5960x and two 2680v3(total of 4cpu)). GPU, DR on 3 Machines(7x 1080 GPU)
It seems that GPU have much less noise, in order to get same amount of noise it doesn’t need to be set at 0.005, but since this is benchmark i left it like that…
I just hope forum wouldn’t kill details with compression.
edit: Forum resized images, so here are the resolutions.
Interior is 2500x2500, Material preview is 1500x1500, Car is FullHD(1920x1080) and in second post GPU 4k(3840x2160)
Thanks for sharing !
The noise level is somewhat fair, but not absolutely. Being too different render engines it is best to compare how the images look at the end.
For multi GPU setups it would help to increase the rays per pixel / Ray bundle size to something like 96/256.
I have notice that noise is much different between those two engines, it actually looks same even with noise 0.01 when using VrayRT still trying to figure all that out. I have try 96/256, it is much better, kpaths are locked, there is no that much oscillations.
You are correct, there is quite difference there, i didn’t have time to deal with it, but camera uses lens profile and other distortions(aniso…) i don’t think it is fully supported scene wasn’t built for gpu in mind, it was just benchmark for raw processing power of GPU’s.
Yes it looks like if we are crossing eyes or drunk too much. But stil l a very good faster preview and then we have to launch production renders on cpu.
