Fixed the nan issue. 6:33.578 with 65536 spp. 256 passes of 256 spp

Path tracer running on the GPU. 4096 samples 16 passes of 256 spp each in 24.524s Also fixed my specular multiplier which was slightly incorrect. The image doesnt converge super well and there are occasional black pixels that appear, with more appearing as the render progresses, so I probably have a NaN somewhere.

1:03 with 512 samples

4:13 with 512x4 samples and 64 bounces. Looks exactly the same

8:26. 512 SPP per pass and 8 passes. 1000 bounces. Not 100% correct but its as close as I think Im gonna get.

And this is with bounces only. Refraction is totally busted and theres a weird shadow on the diffuse ball too.

Decided to just go one part at a time. This is with the light estimation part. Bounces are still not quite working.

getting there

Fixed some issues with what BRDFs and PDFs I was applying where. This is with 1000 passes.

First attempt