Originally Posted by uglykids
A proper rendering was shown in the maxwell forum, showing the caustics after 1 hour at SL21.
Least to say its rather pathetic to draw this faulty comparison, I'd suggest to investigate before claiming and assuming wrong cases.
Maxwell could do these caustics well before 2012.
1) read my disclaimer
2) watch your tone and stay polite, please
3) By "reliably render" I've meant "in most scenes with good quality and acceptable render time". 1 hour for a cornell box is not an acceptable time. Of course it is possible to render such caustics with all unbiased algorithms (excluding them requires extra effort - they are "turned on" by default just by implementing the algorithm correctly). It just takes different amount of time for each one of them. The point is, pre-2012 algorithms (bidir, pt, metropolis implementations without manifold exploration) cannot render scenes with SDS paths reliably - PT won't clean up any caustics, BDPT won't clean up any reflected caustics, PPM won't clean up diffuse parts of the scene, and different metropolis flavors will just randomly succeed or fail depending on a scene and its settings. You can try grabbing the VCM and manifold papers and rendering them with Maxwell. I can guarantee you it will fail horribly in at least one scene, if you don't use either of the two algorithms.