From the bug report data, the GPU that crashed (0000:c2:00.0, device ID 10de:2684) reads as rev ff in config space. When a device reports rev ff and the entire config space reads 0xff, it means the endpoint is no longer responding on the PCIe bus — the GPU is electrically unreachable at the time the report was captured.
The root port for that GPU (0000:c0:03.1, AMD Starship/Matisse GPP Bridge) shows:
LnkCtl: ASPM L1 Enabled
LnkSta: Speed 2.5GT/s, Width x8
A speed of 2.5 GT/s (PCIe Gen1) here likely indicates that the link retrained to the lowest speed after the endpoint stopped responding.
The kernel command line does not include any PCIe power-management overrides:
BOOT_IMAGE=/boot/vmlinuz-6.8.0-49-generic root=UUID=... ro systemd.unified_cgroup_hierarchy=false
There is no pcie_aspm=off or pcie_port_pm=off.
AER status in the report appears clean — no correctable or uncorrectable PCIe errors are recorded on the root ports. That suggests the link itself was electrically stable prior to the failure.
One possible explanation for this pattern is a link power-management transition issue. When ASPM L1 is enabled the root port may transition the link into a low-power state, and some devices fail to resume correctly from that state.
Similar ASPM-related Xid 79 reports have been discussed on the NVIDIA forums, including cases where disabling ASPM resolved the issue:
https://forums.developer.nvidia.com/t/random-xid-79-crashing/358602
Diagnostic step: disable ASPM in BIOS/UEFI
As a diagnostic step it may be worth disabling ASPM in the BIOS/UEFI settings.
On many AMD EPYC and Threadripper platforms this option is typically located under AMD PBS or NBIO PCIe configuration in the BIOS menus.
Common paths look like:
Advanced
→ AMD PBS
→ PCIe ASPM Support
or
Advanced
→ AMD CBS
→ NBIO Common Options
→ PCIe ASPM
Set PCIe ASPM = Disabled, save, and reboot.
After booting, verify the link configuration:
sudo lspci -vvs 0000:c0:03.1 | grep -i "LnkCtl\|ASPM"
sudo lspci -vvs 0000:c2:00.0 | grep -i "LnkCtl\|ASPM"
Both should report ASPM Disabled.
If the crashes stop once ASPM is disabled, that would indicate the issue is related to PCIe power-state transitions.