Anyone see their framework turn off without warning. Specifically when accessing the NvME.
Tried multiple NvME Drives. Samsung and Kingston couldn’t even get through a ubuntu 25.04 install and just turn off. WD_BLACK did but when i used NvME heavily (ie download a larger AI model) it would just turn off.
Is this a full desktop, or did you buy the mainboard on it’s own? What power supply are you using?
I bought the board by mistake and then had to purchase all the pieces independently (from framework) to make a full DESKTOP =(
PSU was not available so i tried a Corsair and Deep Cool PSU (450 and 650 watt) that came from working system. Both experience this problem.
I should also point out i stress tested it using mprime for 15 mins or so, pushing thermals and maxing cpu and it was stable.
Is your SSD WD SN850X? they use a lot of power. I wonder whether it’s a power related issue.
I doubt it is the nvme ssd causing the problem.
It is more likely a RAM problem or a bug in a gpu driver. Running LLM uses the gpu extensively.
You might find it useful searching this forum for FTH and FTR and see if any of those match what you are seeing better.
I’ve had a similar issue with the machine restarting constantly, though it only occurs if it is left at idle. Literally just running an empty process on a single core will keep the system up indefinitely. Or playing a game on the Windows partition. Still a mystery.
So i been debugging with support extensively but really getting no where. Thought i reach out here in case there are any ideas we may have overlooked. Right now i bought a framework PSU since they are in stock and just waiting for it to ship and ill try that 
Started with a Samsung 970 EVO Plus 250GB then for good measure a Kingston NV2 SNV2S/1000G 1TB. In both cases i could not get through a Ubuntu 25.04 minimal install. Around the time it starts touching storage (usually when it starts writing to it) the machine would turn off. Switch to a brand new WD_BLACK SN7100 2TB and it managed to get through the install and even booted the OS. But would still randomly turn off seemingly around hi IO writes. (Any by high last time it died during the setup of apt get of prerequisites for some framework logging tool)
Before i tried the WD Black i managed to get the OS installed on a USB stick and it seemed to work although it was extremely slow for AI.
Speaking of AI it can run a small GPT 20 model no problem, but DOWNLOADING the 120 was a challenge (not running, just downloading) would shut off 1/2 way through the download. However when i got lucky enough tot complete the download it ran fine.
Ran a full run of memetest86 without any issues (setting gpu ram to 512 megs). Even managed to run mprime to test thermals and cpu load without a problem too.
Have you tried nvme devices with firmware (the nvme firmware) updated to the latest version?
You’ve tried multiple nvme devices and I expect support would have suggested doing this; hence, I think it’s unlikely this will help. But if you haven’t tried it, then it might be worthwhile.
Thanks, but yes i checked nvme firmware along with bios etc.
I had the exact same issue, different bios settings would make the problem go away and then it would start coming back, weird stuff. Turns out it was the power supply – I reused an older 500 watt unit. When I ordered a 750 watt Corsair it worked flawlessly.
What I think is happening is that the MB draws less than 400 watt normally, but it periodically has micro surges. Older motherboards had capacitors that would absorb the short bursts, but modern designs rely on the power supply to have the spare capacitance.
(I could be totally wrong about the reason, but on the plus side the power supply fan never needs to come on as it is so overspec’d).
Edit: After rereading your post, my symptoms also matched up with the Samsung SSD. No issues when I switched over to the smaller SSD from my raspberry pi, only with the Samsung. Again, all issues immediately went away with the new power supply. Don’t know if the newer model type fixed it, or if it is because the PS is over spec’d.