Public bug reported:
## Summary
This machine has suffered recurring hard hangs/resets for several
months, with quiet periods lasting weeks or more than a month. On
2026-10-12 it restarted eight times between 01:31 and 02:54 EDT. The
next-boot AMD reset register reports five uncorrected-error data-fabric
sync floods and three hardware-watchdog resets.
One failed boot ends with `amdgpu 0000:0b:00.0: Dumping IP State`,
followed by a sync-flood reset. No component-specific machine-check
record, GPU core dump or vmcore survived. We are reporting the
kernel/platform failure for triage; the faulty component and whether the
initiating defect is hardware, firmware or a driver remain unknown.
Expected behavior: remain running, including unattended or light-load
operation, without a platform reset.
Actual behavior: abruptly stops logging, then reboots; there is no orderly
shutdown in the affected boot tails.
Reproducer: intermittent; no reliable sequence is known. The machine is now out
of production and available for controlled tests.
## Current incident: eight resets
Date: 2026-10-12. All times are America/New_York, UTC-04:00. Each time
is the new boot's beginning, with the hardware reason for the preceding
reset.
- 01:31:20 — 0x02000800, hardware watchdog
- 02:02:08 — 0x02000800, hardware watchdog
- 02:15:45 — 0x08000800, uncorrected error / data fabric sync flood
- 02:20:33 — 0x08000800, uncorrected error / data fabric sync flood
- 02:31:45 — 0x08000800, uncorrected error / data fabric sync flood
- 02:36:19 — 0x02000800, hardware watchdog
- 02:45:46 — 0x08000800, uncorrected error / data fabric sync flood
- 02:53:39 — 0x08000800, uncorrected error / data fabric sync flood
Every retained Oct 12 boot initialization identifies Ubuntu kernel
7.0.0-38-generic. Its package was installed Oct 8, before the overnight
incident.
The last kernel record of the boot immediately before the 02:20 reset
is:
2026-10-12T02:19:51.442825-04:00 amdgpu 0000:0b:00.0: Dumping IP
State
No subsequent "Completed", ring-timeout details or GPU recovery
completion was retained for that boot. The attached sanitized boot log
contains initialization through this final line. Boot-time
`optc31_disable_crtc` REG_WAIT and HDMI vendor-infoframe errors also
recur, as shown in that log.
## Remote desktop and intervention timeline
- A native GDM/GNOME Wayland desktop and amdgpu remained active throughout the
later failed boots.
- NoMachine was started at 02:07, after the first two watchdog resets.
- Its VPN wake watcher was stopped at 02:13.
- An attempted software-rendering configuration caused repeated GNOME Shell
crashes around 02:23–02:24. That failed experiment was removed before GDM
restarted at 02:29. Those application crashes must not be conflated with the
earlier hardware resets.
- NoMachine was stopped at 02:29:05 and masked. Four further system resets
followed at 02:31, 02:36, 02:45 and 02:53. It did not start in the last three
failed boots.
- The last two resets occurred after production services had already been
fenced.
- No persistent NoMachine EGL injection is configured in the native GNOME Shell
service. After evidence collection and graphics updates, the administrator
requested restoration of NoMachine; it was unmasked and started at 09:51 EDT.
NoMachine remains enabled for remote administration, with automatic startup
restored.
## Hardware and software
- CPU: Ryzen 5 7600X, family 25/model 97/stepping 2; microcode 0x0a60120a.
- Board: ASUS TUF GAMING B650-PLUS WIFI, Rev 1.xx.
- BIOS: 3072, build 2024-12-20, AGESA 1.2.0.2b.
- RAM: 2 x 16 GB TeamGroup UD5-5600 single-rank non-ECC; DMI reports configured
5600 MT/s at 1.1 V. EXPO/PBO/Curve Optimizer settings are not yet verified.
- CPU boost is disabled through the existing Linux setting; amd_pstate is
active.
- GPU: integrated Raphael 1002:164e rev c7, subsystem 1043:8877, PCI
0000:0b:00.0. No discrete GPU is currently installed.
- Display: one HDMI display; 1920x1080 modes advertised.
- Ubuntu 26.04 LTS; linux-image-7.0.0-38-generic 7.0.0-38.38; kernel taint 0.
- At failure: Mesa 26.0.3-1ubuntu1, GNOME Shell 50.1-0ubuntu1.1, Mutter
50.1-0ubuntu2.2.
- AMD graphics firmware package 20260319.git217ca6e4-0ubuntu3.1;
amd64-microcode 3.20251202.1ubuntu2.
- Hardware watchdog sp5100_tco is intentionally active with a 60-second systemd
runtime timeout.
- USB Ethernet is used. Onboard Realtek Ethernet/Wi-Fi, audio and Bluetooth
drivers are blacklisted from earlier isolation work; the attached kernel log
preserves the command-line details with identifiers redacted.
## Retained error evidence
- rasdaemon was running with MCE and memory-error tracepoints enabled; its
recorded MCE/memory/PCIe AER tables are empty.
- pstore and systemd's pstore archive are empty. Kdump is ready, but no vmcore
was captured.
- No BERT/HEST/ERST ACPI error tables or detailed EFI HwErrRec records were
found.
- No active devcoredump remains.
- No kernel panic, OOM kill, thermal trip, PCIe AER error or disk I/O failure
was found in the retained overnight kernel records.
- At 01:00, 01:10 and 01:20, retained sysstat samples show about 97% CPU idle,
roughly 14 GB memory available and low disk activity. These samples end before
the first reset; later samples were not successfully decoded.
- All five disks currently report SMART overall passed; NVMe critical
warnings/media errors are zero, and the RAID10 has all four members active. No
DIMM-specific locator is available from the non-ECC memory/error records.
## Longer history
- Local incident notes document abrupt failures in April and May 2026 on
earlier Ubuntu kernels. Earlier configurations included a legacy NVIDIA GPU and
Realtek network devices that are no longer used. The exact reset mechanism was
not captured for every old failure.
- June 1 notes document Ubuntu 26.04/kernel 7.0.0-22 and a headless isolation
configuration removing the desktop stack and blocking display drivers. The
administrator recalls that failures continued during the headless period; an
exact retained crash log from that interval has not yet been recovered.
- Graphics and RustDesk were reintroduced in June. The administrator recalls
periods of improved stability after the Ubuntu upgrade, recurrence around
RustDesk use, and further quiet periods using NoMachine on demand.
- Notes record hardware-watchdog resets on August 5 and the same 0x08000800
sync-flood signature on August 30 and September 1.
- On September 1, persistent NoMachine EGL injection was removed and remote
desktop was made on-demand. The October incident occurred with that arrangement.
## Maintenance after evidence collection
On October 12, signed Ubuntu updates upgraded 12 graphics packages: Mesa
to 26.0.8-1ubuntu0.3, GNOME Shell to 50.1-0ubuntu1.3, and Mutter to
50.1-0ubuntu2.4. GDM was restarted at 09:40:45 EDT, and the new GNOME
Shell process was verified to map Mesa 26.0.8. The kernel, CPU microcode
and AMD graphics firmware were already at the newest candidates in
refreshed Ubuntu repositories. No host reboot or motherboard firmware
update was performed during this work. Long-term stability after the
graphics update is not yet known.
A bounded memtester 4.7.1 test locked 4 GiB and completed one full pass
successfully in approximately seven minutes, with no reported errors.
This was a userspace test of that allocation, not a full-memory boot
test.
ASUS publishes a newer non-beta BIOS (3881/AGESA 1.3.0.1b); it has been
downloaded and vendor-checksum verified, but the administrator deferred
flashing it. Previous firmware-update attempts are part of the
administrator-reported history, without exact prior versions retained in
this report. Tests at stock memory settings and a clean headless boot
are still outstanding.
## Related upstream reports considered
- https://gitlab.freedesktop.org/drm/amd/-/issues/3467 — open Raphael 7950X
iGPU/GNOME ring-timeout report, updated October 6, 2026. It shows recoverable
GPU resets; our platform reset has not been established as the same defect.
- https://gitlab.freedesktop.org/drm/amd/-/issues/5793 — open report containing
both reset-register signatures and failed GPU recovery, but on Phoenix/VCN4
with sustained VA-API decode.
- https://gitlab.freedesktop.org/drm/amd/-/issues/4841 — closed Raphael laptop
hard-reset report, different platform and graphics configuration.
- https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2148735 — unassigned
reset report with a USB ADB/display transfer trigger, not our observed trigger.
Please advise which platform/MCA or amdgpu instrumentation is most
useful when the system resets before rasdaemon, pstore or kdump preserve
a component-specific record. The attached evidence is intended to
support diagnosis, rather than claim a proven duplicate of any report
above.
## Attachments
- kernel-before-0220-sync-flood.sanitized.log — affected boot's kernel log
through the final amdgpu message.
- reset-and-gpu-events.sanitized.log — all eight next-boot reset reasons and
the final amdgpu message.
- platform-and-update-state.txt — incident versions and subsequent graphics
updates.
Hostnames, IP/MAC addresses, serial numbers and volume identifiers have
been removed. Firewall/audit traffic and non-kernel PID1 startup records
are omitted from the kernel attachment; the original exports are
preserved privately.
** Affects: linux (Ubuntu)
Importance: Undecided
Status: New
** Tags: amd-raphael amd64 amdgpu hard-reset resolute
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2170591
Title:
[AMD Raphael / Ryzen 5 7600X] Recurrent hard resets: data fabric sync
flood and hardware watchdog; amdgpu stops at "Dumping IP State"
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2170591/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
