Wanted to let people know the following successes:
success 1:
radeon GPU (with no monitor attached) passed through to a standalone HVM installed via iso (no qubes packages)
Interactive display was still the virtual GPU in one big desktop window
with mesa-opencl-icd installed
Was able to get OpenCL to acknowledge the GPU with the command (using the command “clinfo”)
success 2:
radeon GPU (with no monitor attached) passed through to a template based HVM (I.E with the qubes packages installed) that used a modified “jijna” template.
Display was still the “free” terminal windows in seamless mode
mesa-opencl-icd installed
Was able to get OpenCL to acknowledge the GPU with the command (using the command “clinfo”), while still interacting with qubes via the seamless terminal windows.
actual application test:
found OpenCL wasn’t supported by my application anymore,
Scrapped plan to use OpenCL
New plan is to try to get openGL to use the GPU to render the applications that show up in seamless mode.
Things that might be relavant to using accelerated openGL, but I have not checked out yet are:
- virtualGL
- primus
- Progress reports: streaming OpenGL/Vulkan calls to sys-gui-gpu
Summary: “clinfo” works with GPU passthrough, even in seamless mode. How stable it is under actual application load is unknown.