Using intels open-image-denoise C++ library for post-processing denoising. this is 64-256 samples, rendered and denoised in 262 seconds. Using caustic map for caustics

yee ole teapot included. rendered in 1080p, adaptive samples 256-1024, 10 bounces, 11 minutes to render, ouch

Looks interesting if you remove the point light but keep the emissive sphere. This is essentially only the global illumination factor

Light sphere intersection, emissive contribution, and added some textures to spice things up a bit

Adaptive sampling with t-test variance. 256 to 1024 samples, 116 seconds, much better.

various optimizations reduced render time to 332s which is a significant improvement. Glossiness needs some work though, I think

Painfully long to render with 1024 samples, 100 maximum bounces, 0.98 rr continuation chance. 1370 seconds... Lots of optimizations to come!

monte-carlo random samples, more noisy for the same pixel count, shoutout antialiasing

stratified perpixel sampling, 16 samples per pixel

Minimum of 10 bounces, 100 max bounces with rr. Not much of a difference

3 min 8 max bounces with rr, 32 samples per pixel. Result is noisy but doable. 39s render time

First render, why is she upside down?!