We are nearing completion and the last phases of development/testing of the new renderkernel for finals we have been working hard on.
We decided to share some images/comparisons here so people can see our progress and what to expect in the upcoming beta 2.5.
Here are two images of a complex interior (scene by enricocerica), which both have 10.000 samples/px.
One rendered with our new kernel named 'PMC' (our custom mutating unbiased kernel written for GPUs) and one with the good old pathtracing kernel.
Both have the same depth, being 32 bounces, and the same shared settings (russian roulette etc)
Both kernels run at approx the same speed in Samples/Sec vs Mutations/Sec.
Both have 10.000 Samples/px and rendered for 1:24 (one hour and 24 minutes) on one GTX460 GPU.
Both images are exact output, unfiltered/post processed, both rendered in 1280x720 (720p).
As you can see the difference is enormous, the fireflies you can see in the pathtraced image (unexplored caustics) are gone and all caustics are propperly explored in the PMC image.
You can also see the complex caustics on the ceiling above the dining table propperly explored.
We still have not optimized the new kernel so it can only get faster/better from here, and if our hopes are correct, we could expect a nice speed gain by optimizations yet.
- Pathtracing kernel
- PMC kernel
so when both are fully done, you can finally render with the PMC kernel and render complex interiors and other scenes with caustics without those nasty fireflies
EDIT
Here's another interior rendering of the scene, with interior area lights only (blackbody), and IES file distributions on all area lights.
This is a very hard scene to render, it's virtually impossible with pathtracing.
This one is the new PMC kernel, with maxdepth = 32 bounces, a PNG image, non-postprocessed output straight from octane.
- Interior PMC kernel with IES/area lights
wow amazing and thanks for sharing