Re: Luxcore direct light
Posted: Fri Aug 09, 2019 9:37 am
First of all something about all the direct light tests I have seen recently and I'm sure you are not aware: when you set max. path depth to 1, BlendLuxCore exports a value of 2 (I think it always exports "n + 1").
Why ? Because so you can have MIS and MIS is pretty mandatory to render glossy surfaces. So it is a wise behavior and BlendLuxCore tries to not let you shoot in your foot.
However, this has clearly a huge impact on the direct-light-only benchmarks you are trying to do lately. LuxCore traces 2 times the rays Cycles does. Indeed, I do my tests outside Blender, setting correctly the max. depth values.
Now this is a 128 samples/pixel done with Cycles:
and this a 128 samples/pixel done with LuxCore (with a max. path depth of 2, like exported by BlendLuxCore):
I would say that the amount of noise is exactly the same (very low).
Now, let's repeat the same test with LuxCore and really use a a max. path depth of 1:
Same result tracing less rays but it is not a surprise, there are no glossy surfaces at all.
If you are bench marking noise, use the same samples/pixel. If you are bench marking performance use wall clock. Don't mix stuff (noise and time).
Why ? Because so you can have MIS and MIS is pretty mandatory to render glossy surfaces. So it is a wise behavior and BlendLuxCore tries to not let you shoot in your foot.
However, this has clearly a huge impact on the direct-light-only benchmarks you are trying to do lately. LuxCore traces 2 times the rays Cycles does. Indeed, I do my tests outside Blender, setting correctly the max. depth values.
Now this is a 128 samples/pixel done with Cycles:
and this a 128 samples/pixel done with LuxCore (with a max. path depth of 2, like exported by BlendLuxCore):
I would say that the amount of noise is exactly the same (very low).
Now, let's repeat the same test with LuxCore and really use a a max. path depth of 1:
Same result tracing less rays but it is not a surprise, there are no glossy surfaces at all.
If you are bench marking noise, use the same samples/pixel. If you are bench marking performance use wall clock. Don't mix stuff (noise and time).