Re: Luxcore vs CyclesX Benchmark
Posted: Tue Feb 01, 2022 10:49 pm
It's not about scrambling, it's about using noise threshold for adaptive sampling. I notice in your video that you actually turned it off.
That's an optimization which brings afaik no downsides, all it does is that it stops wasting samples on places with little noise and instead spends more time on sampling noisy areas. This drastically speeds up rendering in scenes which contain both parts which need few samples (light areas with lots of direct light) and parts which need a lot of samples (areas illuminated by indirect light). It does so by measuring local noise and stopping rendering once it falls below some threshold.
I have scenes where some areas need 4000+ samples to stop being super noisy and other areas are clean with 512 samples. Without the noise threshold checked I have to render the whole scene with 4000+ samples, while using the noise threshold makes the rendering about 3x faster. And that's just using the default Blender setting without changing anything.
Your point about Cycles lacking some features and providing less accurate global illumination is of course correct and important to say, but a large part of your video is about speed as well and turning off this optimization, which brings one of the biggest speedups in Cycles X with no effect on GI accuracy, is imo a bit similar as not using PhotonGI in LuxCore.
That's an optimization which brings afaik no downsides, all it does is that it stops wasting samples on places with little noise and instead spends more time on sampling noisy areas. This drastically speeds up rendering in scenes which contain both parts which need few samples (light areas with lots of direct light) and parts which need a lot of samples (areas illuminated by indirect light). It does so by measuring local noise and stopping rendering once it falls below some threshold.
I have scenes where some areas need 4000+ samples to stop being super noisy and other areas are clean with 512 samples. Without the noise threshold checked I have to render the whole scene with 4000+ samples, while using the noise threshold makes the rendering about 3x faster. And that's just using the default Blender setting without changing anything.
Your point about Cycles lacking some features and providing less accurate global illumination is of course correct and important to say, but a large part of your video is about speed as well and turning off this optimization, which brings one of the biggest speedups in Cycles X with no effect on GI accuracy, is imo a bit similar as not using PhotonGI in LuxCore.