That obscure flag is:
debug-value = 256
That obscure flag is:
debug-value = 256
This has nothing to do with OpenCL versions - Cycles is using openCL 1.2, which to my knowledge is fully supported by NVIDIA. LuxCoreRender and ProRender demonstrate that OpenCL on NVIDIA doesnât have to be slow either.
I really donât like to see this thread turning into endless argument.
For users who recently switched to RTX cards, set the preference to CUDA and use the latest Blender software, it should fix the performance issue.
Enjoy
So then is the performance drop when using opencl with nvidia cards strictly a bug? I donât really use prorender or luxcore, so i wasnât even aware nvidiaâs opencl performance was better in those engines.
Donât worry skw develops for cycles, so he does know more when it comes to whats a bug and whats a hardware issue. I wouldnât start any argument concerning that with him.
I would guess that itâs a matter of optimization. Since Cycles+CUDA works well and stable, there is little incentive for anyone to work on OpenCL/NVIDIA. It would take weeks if work and at best, there would be no tangible benefit for the user.
If forcing nvidia cards to use opencl isnât necessary for using them to render a scene alongside amd cards, then the only other benefit I can think of is an ideological one, where you donât want to use an api that is vendor locked.
As youâve said, cuda works well, so there isnât really a reason to give opencl optimization on nvidia cards more priority over other improvements atm.