Frequent, unrecoverable system freezes with Vega 64 in OpenGL, OpenCL, and other high-power draw workloads.
System:
- Operating System: Fedora Linux 38
- KDE Plasma Version: 5.27.5
- KDE Frameworks Version: 5.107.0
- Qt Version: 5.15.9
- Kernel Version: 6.3.7-200.fc38.x86_64 (64-bit)
- Graphics Platform: Wayland
- Processors: 16 × Intel® Core™ i9-9900K CPU @ 3.60GHz
- Memory: 62.7 GiB of RAM
- Graphics Processor: AMD Radeon RX Vega
- Manufacturer: Micro-Star International Co., Ltd.
- Product Name: MS-7B17
- System Version: 2.0
I'm using a 2560x1440 display @ 165hz over Displayport, but I've also tested with a 1080p 60hz display over HDMI, and it makes no difference. Also, my system has a 750w PSU, so I doubt that is part of the issue either.
The issue - findings through testing:
In games..
Vulkan is more stable, with infrequent, or no crashes. OpenGL is more likely to cause the system to freeze, but only with some applications. Possibly due to the version being used? I wouldn't know.
I can reproduce the crash repeatably and reliably, as I have a Minecraft (OpenGL) instance that triggers a full system freeze as soon as the game is rendered. Interestingly through, when running through a Zink wrapper (env MESA_LOADER_DRIVER_OVERRIDE=zink), the crash no longer occurs.
Other workloads:
GPU accelerated Stable Diffusion (OpenCL) running through https://github.com/AUTOMATIC1111/stable-diffusion-webui will produce the system freeze.
Other details:
Even at idle, the GPU Tach is lit up mid-way, while when on Windows using the AMD driver, it would only light up one LED. I'm not sure how relevent that fact is, but it could be indicating an unnecesarily high power draw.
I also suspect that all of these issues are somewhat related to #716 but this occurs regardless of whether the applications are using vsync or not.