Lighthouse 2 RTX / Path Tracing benchmark

The mighty RTX 6000 has fallen. Strangely enough two RTX 2080 Ti's beat it; I don't think they are overclocked either. Driver issue perhaps? I do wonder now how the Titan RTX perrforms.
 
Possibly driver issue lacking optimizations. In the benchmark do we know RTX is on or off, can you toggle the feature?
I was looking at the OctaneBench benchmark and noticed with RTX enabled the performance increase was about 190%, though the driver is likely optimized.
 
The mighty RTX 6000 has fallen. Strangely enough two RTX 2080 Ti's beat it; I don't think they are overclocked either. Driver issue perhaps? I do wonder now how the Titan RTX perrforms.
If this benchmark can actually utilize more than one RTX card at a time then how would two 2080ti's not beat a 6000?
 
I'm on certified Quadro ODE drivers, vs. the latest and greatest consumer drivers, and i'm probably not even on the latest qualified drivers either.. ( yup, i'm not 441, vs latest is 442)
I think also the 2080 Ti's have a faster clock speed?

My dev system is all about stability, not max perf.
 
Cool. Your video has a lot of white form 1:35. Editing mistake (left rendering to end of audio)?


The noise is temporally stable which I found curious. Is there a particular purpose to that?
 
Cool. Your video has a lot of white form 1:35. Editing mistake (left rendering to end of audio)?
The noise is temporally stable which I found curious. Is there a particular purpose to that?

The video is actually from the older version of the benchmark, and it indeed has an editing mistake.
In the new benchmark the noise is no longer stationary. I find the stationary noise easier on the eyes, but a lot of people find it very distracting.
 
I definitely find it distracting! ;) You also get better temporal averaging, so the overall quality is better with 'cinematic noise', although the contrast of the noise is far stronger than film and I guess pretty fuzzy looking in motion. Highlights the need for various denoising strategies as there's no way we'll be getting enough rays to sample enough to reduce noise to something bearable. I'm guessing 16 samples per pixel is needed for the realm of cinematic noise? I suppose that depends on the contrast of the samples, and HDR scenes will need even more than that.
 
Back
Top