Freezing up under load. Trouble installing Ubuntu. Updated Bios, but now won't post

Framework Desktop Linux
1

Forgive me for using an LLM to summarize this. Its a long mess of a sequence of events and it’s helped me keep track of what I’ve done.

1. Environmental Setup & Baseline

  • Machine: Framework Desktop (strix), featuring an AMD Ryzen AI Max / Strix Halo class processor with 128GB unified LPDDR5 memory.
  • Storage: Single 1TB WD_BLACK SN7100 NVMe SSD.
  • OS Environment: Upgraded from Ubuntu Desktop 2025.10 to 2026.04 (running Linux Kernels 7.0.0-15 and 7.0.0-22). Operating primarily as a headless local LLM compute node.
  • Network/Management: Managed via an external Sipeed NanoKVM (IP KVM) and powered via a Kasa smart Wi-Fi switch.

2. Phase I: Compute Instability & Kernel Panics

While handling a local LLM runtime execution (qwen3.6-27b-rocmfp4 inside Docker via ROCm), the system suffered severe AMD GPU compute queue hangs:

  • The vllm-strix-qwen36-awq-agent container exited with code 139 (segmentation fault).

  • The host kernel began logging widespread amdgpu driver timeouts: amdgpu ring comp_1.1.0 timeout, MES failed to respond, and failures to unmap/reset legacy queues.

  • The driver failure cascaded into memory management corruption, throwing a BUG: Bad page state and a __rmqueue_pcplist allocator crash.

  • Subsequent boot attempts on the primary kernel threw systemd PID 1 failures (Protocol error when forking sandboxing generators). Rolling back to the 7.0.0-15 kernel reached a TTY terminal, but quickly locked up upon user interaction.

3. Phase II: Failed Rescue & Installer Corruption

  • The host was successfully booted into rescue.target with amdgpu blacklisted. Docker services and sockets were masked to prevent accidental container activation. The system stabilized temporarily over SSH at 192.168.50.11.

  • An attempt to completely wipe and reinstall the OS via an Ubuntu Server 26.04 USB installer stalled indefinitely during the curtin extract / rsync phase.

  • Post-stalls diagnostics revealed that while the NVMe target was healthy, the USB installer media itself was heavily corrupted—failing multiple squashfs MD5 checksum verifications and causing Python 3.12/libapt-pkg segmentation faults and __rb_insert_augmented kernel null pointer dereferences.

4. Phase III: The Firmware Event & Current No-POST State

  • During the recovery cycle, the NanoKVM captured an automated Framework BIOS update routine attempting to migrate firmware from version LFSP0.03.04 to LFSP0.03.05.

  • The screen confirmed a successful memory test phase (137,438,953,472 bytes of system memory tested OK), but the screen abruptly cut to black shortly after. Power may have been inadvertently interrupted via the remote Kasa switch or an internal firmware-initiated reboot sequence failed.

  • Current Symptoms: The machine completely fails to POST. There is zero video output via HDMI/KVM, no DHCP reservation, no local network ping, and Wake-on-LAN (WOL) attempts are unresponsive. The physical status indicators show a spinning fan, a solid white light next to the power button, and two solid red lights internally on the mainboard (one near the NVMe harness, one at the bottom-right of the fan assembly). CMOS reset, NVMe removal, and Wi-Fi card extraction have yielded no changes.

Edit:
BIOS release channel:
I did not download an Alpha/Beta firmware from the Community Forum, and I did not manually download a BIOS package from the Knowledgebase. The update was offered through LVFS/fwupd on Ubuntu. I would classify it as the LVFS/fwupd-provided stable firmware unless Framework’s LVFS metadata says otherwise.

BIOS version:
The updater screen showed:

BIOS: LFSP0.03.04 -> LFSP0.03.05

Update method:
LVFS/fwupd using fwupdmgr from an Ubuntu live installer environment. It was not the Windows .exe method, and it was not a manually downloaded UEFI BIOS update package.

Optional extra detail:
The Framework firmware updater screen appeared after staging the update and showed the AC power warning plus successful memory test output before the machine later went black/no-POST.

2

I’d start off with removing the KVM and rescuing with the computer on your desk and a monitor and keyboard plugged in.

Can you get into the BIOS after the memory test phase?

3

nope. There’s no output via HDMI. I’ll have to test with a display port when I get the chance. also: BIOS release channel:
I did not download an Alpha/Beta firmware from the Community Forum, and I did not manually download a BIOS package from the Knowledgebase. The update was offered through LVFS/fwupd on Ubuntu. I would classify it as the LVFS/fwupd-provided stable firmware unless Framework’s LVFS metadata says otherwise.

BIOS version:
The updater screen showed:

BIOS: LFSP0.03.04 -> LFSP0.03.05

Update method:
LVFS/fwupd using fwupdmgr from an Ubuntu live installer environment. It was not the Windows .exe method, and it was not a manually downloaded UEFI BIOS update package.

Optional extra detail if they ask:
The Framework firmware updater screen appeared after staging the update and showed the AC power warning plus successful memory test output before the machine later went black/no-POST.

4

have you tried to remove battery for 1 minutes and remove power cord? it resets EC, maybe it will help

5

Not yet. Funny I hadn’t tried it yet