Anybody else experiencing low-power instabilities? My experience on Cachy, Fedora, or Windows is identical to what the user is experiencing in this Reddit thread. Left idle at ~5-7 W power draw, the system will eventually restart (almost always <8 hours since last use). Setting max C-states to 1 via grubby and confirming only C1 available via powertop does not fix the issue, it also results in an eventual restart. The PC will not restart if leaving a sustained load of any kind, to include an indefinite empty while loop on a single thread or keeping a game open.
All OS’s are up to date, all drivers are up to date, I installed the 3.03 FW BIOS. Current kernel on Fedora 43 (6.17.7-300.fc43.x86_64).
journalctl logs terminate abruptly with no error, it’s like the machine is physically unplugged from the wall and loses all power instantly. Windows Event Viewer reports generic loss of power, Event 41. This is a batch 10 machine. This has been an issue since day 1. I disassembled and reassembled the PC to ensure everything is seated properly, unlocked additional power settings in the Windows registry to mimic the max C-state = 1 setting in grubby to have an idle power draw ~14-20 W. I’ve also tried disabling PCIe ASPM and other power saving settings. And again, zero issues when there is a sustained load even if I restore PCIe ASPM and the deeper C-states on both Fedora and Windows.
CPU is AMD RYZEN AI MAX+ 395 w/ Radeon™ 8060S × 32 with 128 GB RAM
Any other ideas? Maybe just a bad PSU? Simply a bug I need to wait for a BIOS update for?
I did a bit more investigating on the Fedora side. It might be a thermal issue. I ran a while true loop on a single thread while on Performance mode, Balanced mode, and while in Power Saver mode. There is an immediate spike to ~30 W and 100 degrees C on the thread running this empty process while in Performance or Balanced modes. In (edit) power saving mode, the power draw drops to ~15 W and ~40 degree temps. This spike and these high temps do not occur in games on Windows, they’re usually in the 60-70 degree territory then. Do you think it might be the case that one of these background processes triggers a boost clock that superheats a core and triggers the emergency shut off? Being in power saver mode is OK with me in Fedora, I have other needs for the RAM than LLMs and the performance hit is fine. But I do need Performance mode to work in Windows for games. I’ll keep digging.
I previously disassembled, repasted, reseated, and tightened the heatsink to ensure it had good mounting pressure. From factory, I noticed the paste application was pretty light. I’ll triple check this.
Alrighty, I’m at a ‘stable’ workaround state but the PC cannot sustain long true idle durations. It always restarts, always within a few hours. The error is a sudden power loss and is never reported in system logs in Fedora, only registers as generic Event 41 in Windows. The workaround is to throw away a thread on login to run the empty process indefinitely. Total system power use is in the range of 10-14 W, slightly above ambient temperatures now.
I resolved the thermal issue by repasting again and cutting some of the heat conductive tape around a section that was raised really high by a small chip. I noticed the imprint of the RAM chips in one of the thermal pads was quite shallow in comparison to the others. I was hoping this modification would improve the contact all around, seems to have done so. I can now run the while true loop indefinitely while in Performance mode at the same power draw and thermals as in Power Saver mode.
Of course, grain of salt, but AI suggested a low-load power instability issue involving the PSU or VRMs on the mobo. Frankly, I don’t know how I would go about diagnosing if either of those are the case.
Thank you for the suggestion, I downloaded Ryzen Master on my Windows partition but the new APUs don’t appear to be supported at present.
Put in a support ticket, I’m at a loss of what else I can do. Running an empty process on a single thread lets the PC at least be usable but it’s not ideal. Does suck that the thermal issue reappeared though but the peak temps are down to ~90 degrees now instead of ~100 degrees, so… progress? I’m just running in power saver mode where the background while true process runs at 15 W and ~40 degrees.