LuxCoreRender Forums

My Ryzen has no OCL driver and will only work natively, so I cannot test the original configuration of the file.

About OCL on AMD, I used this library once and was able to use OpenCL on AMD Ryzen, not sure if it works with your machine but you can give it a chance. http://portablecl.org/

TAO wrote: ↑Wed May 13, 2020 3:13 pm About OCL on AMD, I used this library once and was able to use OpenCL on AMD Ryzen, not sure if it works with your machine but you can give it a chance. http://portablecl.org/

Thanks for the link, but I'm no developer and don't know how to install or build libraries
In general CPU native rendering should be faster and using OCL on CPUs should be avoided, or so it is told somewhere in the docs, don't know where right now.

Back to your GPU problem, it is not with Lux and not in the scene, then it might be something related to your system.
Does the GPU work as expected in Cycles and Luxmark?

I think thats gonna answer your question.

This looks just alright, now that it is based on LuxCore 2.2 makes it very strange that you cannot achieve similar GPU performance with the very same v2.2 LuxCoreUI and your simple scene
Sorry to be of no help, but this problem needs smarter heads than mine to solve

I find the same very strange behaviour with TAO's scene, i.e. GPU is very slow, with LuxCoreUI 2.2 (but 2.4 is ok).
This line in the config file seems to be the source of the problem, at least on my laptop: If I comment it, then GPU speed is again very high.

acasta69 wrote: ↑Wed May 13, 2020 9:03 pm I find the same very strange behaviour with TAO's scene, i.e. GPU is very slow, with LuxCoreUI 2.2 (but 2.4 is ok).
This line in the config file seems to be the source of the problem, at least on my laptop:

Code: Select all

opencl.gpu.workgroup.size = "0"

If I comment it, then GPU speed is again very high.

"0" means use the work group size suggested by the driver (and often the drivers can be very wrong). The value should be 32 for NVIDIA and new AMD, 64 for some old AMD GPU.

You are great Dade, Not Only that fixes the issue but fixing this also fixes 3dsmax behavior too, and now it's much faster.
Thank you all.
Is there any suggestion for CPU workgroup size ?

Last time I checked, OpenCL CPU device was much slower than native C++, so I wouldn't use it at all (set opencl.cpu.use = 0).

I used opencl.cpu.use = "0" but it's not work exept i add 0 for cpu to opencl.devices.select = "10".
Anyway, it's work much better now.
Thank you.