On March I have moved 100% from Windows to Manjaro. AMD does not provide "official" drivers for Manjaro, but there are community ones.
I am a big fan of providing computing power for scientific purposes via Folding at Home and BOINC. I have two eGPUs - one with RTX 2000 Ada and the second one with Radeon Pro W7500.
Setting up Radeon was harder from the start, but for a few weeks it was running Folding at Home correctly. I had been using ROCm at that time. I remember that it did not work completely without ROCm.
Unfortunately, on 30 August I updated everything on Manjaro, and the next day Radeon stopped working and I can’t repair it till today. That day kernel updated from 6.12.41 to 6.12.44, but there were also other updates, linux-firmware etc.
the Radeon was working correctly, when Folding at Home saw it as “gfx1102” in web client
I tried reinstalling ROCm multiple times, reinstalling linux-firmware, amdgpu-firmware, amdgpu-pro, practically reinstalling or installing everything what Gemini and GPT knew; downgrading them to previous versions; today I even tried to switch Kernel from 6.12 to 6.1 as Gemini and GPT were saying to me, that kernel 6.12 is too new and therefore not fully supported by AMD drivers and with 6.1 it would be nice - it ended up with system not booting but doing memtest all the time, but at least for the first time in my life I was able to repair the grub with Live USB so I’m happy that I tried and learned something. I tried also cloning some packages from git and building them on my PC, I tried adding some environmental variables, nothing worked.
With some of these software I can get FAH and OpenCL “see” Radeon, but OpenCL “does not connect” with Radeon - clinfo displays number of devices 0, FAH displays supported: “false”. I also tried repeating the steps I did in early August when I was able to set it up, but I was unable to get the same results now.
“Best” I was able to do in last 9 days, was installing opencl-mesa (aka rusticl), and with that Radeon is detectable, supported: “True” by FAH and connects with opencl, but it crashes many working units before getting stable one, then gets joint working unit with Nvidia GPU, and software like radeontop and sensors suggests, that Radeon is like getting vestigial work, it never gets upon 20 Watts; it seems like Nvidia is doing two tasks at once with ~50% power on each (one solo, one joint with Radeon), and Radeon is theoretically working but doing almost nothing. When I dont connect Nvidia GPU, Radeon is in loop of downloading tasks and crashing them infinitely.
When it comes to BOINC, it never worked with Radeon - it was downloading tasks but they were crashing after 2 seconds, no matter which project. With opencl-mesa it does not even download them, downloading tasks worked only until 30 August when Radeon was displayed as “gfx1102” by FAH.
All cables, eGPUs, PSUs etc. are ok, the problem is 100% software.
I would be very grateful for any advice, By the way, AMD providing better software support for gaming GPUs than “professional” GPUs on Linux is irrational. Just like I mentioned before, if You tell me what to do so FAH sees my Radeon as "gfx1102" it will be 99% enough for it to work.
Update:
listings from my console:
sudo mhwd -a pci free 0300
Skipping already installed config ‘video-linux’ for device: 0000:2e:00.0 (0300
Skipping already installed config ‘video-linux’ for device: 0000:06:00.0 (0300
Skipping already installed config ‘video-linux’ for device: 0000:00:02.0 (0300:8086:a7ad) Display controller Intel Corporation Raptor Lake-U [Intel Graphics]
sudo mhwd -a pci nonfree 0300
Skipping already installed config ‘video-linux’ for device: 0000:2e:00.0 (0300
Skipping already installed config ‘video-linux’ for device: 0000:06:00.0 (0300
Skipping already installed config ‘video-linux’ for device: 0000:00:02.0 (0300:8086:a7ad) Display controller Intel Corporation Raptor Lake-U [Intel Graphics]
inxi: https://pastebin.pl/view/cdfe9046