AMD GPU tuning

[JLemur]

Hi,

Could anyone help me understand if there are tuning parameters I can apply to the amdgpu rocr driver so that OpenCL will work as well in Linux as it does in Windows on the same laptop? In Windows, I'm getting a Geekbench OpenCL score around 8,000, but in Linux Mint it's less than 3,000. I'm hoping to use OpenCL to accelerate Darktable, but when I have OpenCL installed, darktable states that the GPU is too slow and it disables the GPU. If I run Darktable in Windows, it enables the GPU with a "very fast GPU" profile.

I noticed in clinfo output that Linus states I have 11 compute units, but Windows states 10 computes units. I'm only expecting 10 since it is a "Radeon RX Vega 10" integrated graphics. Even if I try another distro, like Ubuntu or OpenSUSE, clinfo states 11 compute units. There are some other things that don't match up between Windows and Linux clinfo, like cache size, which I'm wondering if it explains why Windows is operating better than Linux.

[JLemur]

Here is the darktable-cltest output in LInux. I couldn't attach it initially due to file size, but I pared it down to what can be attached.

Windows OpenCL Geekbench OpenCL results: https://browser.geekbench.com/v5/compute/5485862
Linux OpenCL Geekbench OpenCL results: https://browser.geekbench.com/v5/compute/5486497

I installed the amdgpu via the amdgpu-install --usecase=opencl --opencl=rocr.

This is the version of amdgpu-install: amdgpu-install_22.20.50200-1_all.deb

[SMG]

Could anyone help me understand if there are tuning parameters I can apply to the amdgpu rocr driver so that OpenCL will work as well in Linux as it does in Windows on the same laptop?

The rocr driver is part of the proprietary AMD driver. You might want to ask this question on the AMD community forum.