Small update. I downgraded to FW 3.03 and I see the same issue also with this FW. I might try to downgrade to an older kernel if I have time.
So, I’ve been trying some more combinations and compiled the kernel with the patch that is floating around. Turns out the same issue can be triggered by pure CPU load.
I’ve also downgraded the kernel and I still hit the isse.
@Tom5 can you load a model and hit it with this GitHub - eugr/llama-benchy: llama-benchy - llama-bench style benchmarking tool for all backends · GitHub ? I’m interested to see if this works stable with your setup. I tried to replicate it but for me it crashes.
Hi @Lafunamor , just tried benchy und it succeeded (finished quickly - doesn’t seem to produce high load):
| model | test | t/s | peak t/s | ttfr (ms) | est_ppt (ms) | e2e_ttft (ms) |
|---|---|---|---|---|---|---|
| Qwen3.6-35B-A3B-MTP-UD-Q4_K_XL.gguf | pp2048 | 799.87 ± 6.33 | 2144.96 ± 13.47 | 2144.42 ± 13.47 | 2144.96 ± 13.47 | |
| Qwen3.6-35B-A3B-MTP-UD-Q4_K_XL.gguf | tg32 | 51.90 ± 7.58 | 53.45 ± 7.81 |
llama-benchy (0.3.8.dev2+gff162bcfc)
date: 2026-05-12 18:41:14 | latency mode: api
Thanks @Tom5, for checking! Meanwhile, I’ve opened a dedicated issue on freedesktop about my problems and the friendly AMD staff analyzed my data and suspect a thermal or hardware issue with my unit. I’m in contact with framework support and asked if I can get the board exchanged. So it seems that I just had bad luck ![]()
You are welcome @Lafunamor. Maybe that you are the unlucky one with a hardware defect. My fingers are crossed - please keep us updated.
Btw. I saw at least one case report in the forum, describing temperature spikes due to a CPU cooler assembled without removing the protective sticker beforehand. I hope that you are not affected by this issue.
I’m glad some of you were able to resolve your issues and get more stable systems but to throw in my $0.02 you can count me among those seriously regretting the purchase. Maybe I also just got defective hardware (which itself doesn’t inspire confidence, if all the people on these forums having issues got lemons there are some serious QA issues at the factory) but a big part of the problem with AMD’s lagging and frankly terrible software support is that you can’t really tell if it’s a driver or hardware issue until you’re out of warranty. I’m well aware of their track record of not properly supporting their cards until a year or more after release, so I initially assumed the issues I was facing would be solved eventually by software updates. Now that I’ve been sitting on a $3k paperweight for almost a year I’m starting to think I should have just returned it day one.
Fwiw I work in IT as an engineer so I’m very comfortable troubleshooting but it’s not something I want or should have to do in my personal life with such an expensive piece of consumer hardware. As it stands I wouldn’t recommend the Desktop to anyone who wants a functional machine or isn’t interested in hours of debugging something they shouldn’t have to debug.
Just finished putting together my custom build with the 128GB variant mainboard, thought I’d share my experience so far. Overall stability has surprisingly been great, I’ve had just a single freeze up while gaming. I’m running windows 11, and grabbed the official driver package from framework, before running AMD adrenalin and updating everything there (I believe it updated the graphics driver and chipset). I don’t know if the general instability is mainly being seen on the Linux side, but my system has been working as wanted.
Edit: should also clarify that I’m using the most recently released BIOS.
It has been relatively stable. But it is not flawless like something such as ThinkPad. Sometimes I experience minor freezes, and even USB-C extension card issues that I have to power off for 5 min to solve.
I’m on Fedora. So is it stable? Yes relatively, but not workstation grade for sure.
I am running my Framework Desktop with Linux Mint 22.3 with the xanmod kernel to keep up with the latest stable kernel for the latest hardware support. The only issue I have had has been issues with power management and failures when waking from sleep. From the start, various devices will fail to wake after resuming from sleep. This was mitigated with some GRUB Command line parameters and some udev rules to prevent the USB ports from powering down when the computer sleeps
# Added amdgpu.dcdebugmask=0x610 to disable Panel Self Refresh(PSR) to prevent computer from freezing when it sleeps.
# 0x10 = Disable PSR (Panel Self Refresh)
# 0x400 = Disable Panel Replay
# 0x200 = Disable Seamless Boot# Added ttm.pages_limit=28835840 ttm.page_pool_size=28835840 to use GTT to use system memory as Video memory
# Added amdttm.pages_limit=28835840 amdttm.page_pool_size=28835840 to use GTT
# Added amd_iommu=pt lower latency# Added amdgpu.runpm=0: Disables the aggressive Runtime Power Management that puts the physical DisplayPort PHY links into an unrecoverable deep sleep sleep state.
# Added amdgpu.gfxoff=0: Prevents the driver from letting the graphics core go completely black when idle, ensuring the link protocol layers remain receptive to wake calls.
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash amdgpu.dcdebugmask=0x610 amd_iommu=pt ttm.pages_limit=28835840 ttm.page_pool_size=28835840 amdttm.pages_limit=28835840 amdttm.page_pool_size=28835840 amdgpu.runpm=0 amdgpu.gfxoff=0"
The latest issue I have had started with the 7.0 kernel. Now, when my monitor goes to sleep, the DisplayPort won’t resume sending the video signal until I drop to a terminal (Ctrl+Alt+F3) and run some commands to reset the link,
export DISPLAY=:0
xrandr --auto
DISPLAY=:0 xrandr
then go back to the GUI display (Ctrl+Alt+F7)
I have been using one for a few months now. Dual booting W11 and CachyOS on two different M.2s. Overall it has been very stable. I did some light gaming on it but am looking to sell it and get a more powerful machine dedicated to gaming. I was impressed by how quick it was for its physical size though.
If there was going to be a fix for these stability issues, would it come from the linux side or the AMD/BIOS side?
Thanks for the feedbacks.@Alfred_Simkin which stability issue do you mean? In the meantime I did upgrade to Kernel 7.0.9 (after 7.0.8 and 7.0.4) and haven’t experienced any issues so far.
That’s good to know. I was referring to the random reboots, video issues, and graphics card problems that people referred to at the beginning of the beginning of the post. I’m glad to know those seem to be fixed in the new Kernel.