Recurring SIGILL crashes in HA Core (python3) after migrating to new NUC13 — hardware ruled out via memtest, still unresolved

Summary

Since physically migrating my Proxmox host to a new NUC13i5ANH, Home Assistant Core, which is running in it own VM has been crash-looping with an illegal-instruction fault (SIGILL / signal 4) roughly every 15 minutes to 2 hours, sometimes longer. The VM itself is stable, it’s just HA that keeps restarting. The Supervisor watchdog catches it and restarts Core each time, so the system self-heals, but this has been happening continuously for several days now. I’ve exhausted every fix I can think of and I’m hoping someone has seen this pattern before.

Hardware / Environment

  • Host: ASUS NUC13ANBI5 (13th Gen Intel Core i5-1340P, 4 P-cores + 8 E-cores, hybrid architecture)
  • 32GB RAM (2× 16GB DDR4-2400)
  • Samsung 970 EVO Plus 250GB NVMe
  • Proxmox VE 8.4.17
  • Guest: Home Assistant OS 18.1, Core 2026.7.2, running as a QEMU/KVM VM (4 vCPU, 16GB RAM, OVMF/UEFI, q35 machine type, VirtIO disk/network)
  • BIOS 0044 (latest available), microcode revision 6134 (also latest)

The error, every time, identical:

systemd-coredump: Process XXXX (python3) of user 0 terminated abnormally with signal 4/ILL

Corresponding Supervisor watchdog log shows Core exiting with code 132 (128+4=SIGILL) at the same timestamp, 1:1.

What I’ve tried — all unsuccessful:

  1. Cold VM stop/start (to rule out stale CPUID caching)
  2. Three different Proxmox CPU Type settings: host (full passthrough), x86-64-v2-AES, x86-64-v3 — all produced the identical crash pattern at similar frequency
  3. Physical RAM reseated + Memtest86+, 2 full passes on all 32GB — 0 errors
  4. NVMe SMART health check — clean (no media errors, no critical warnings)
  5. Fresh Home Assistant Core container rebuild (ha core rebuild) — crashed again ~15 min later
  6. Full BIOS/microcode update (0038 → 0044, microcode 4124 → 6134) — crashed again within hours
  7. Built a second, completely fresh HAOS VM from the official OVA image, restored from backup, moved all peripherals over — same identical crash, twice, a few hours apart — this ruled out anything specific to the old OS install/filesystem state
  8. Disabled a local Whisper STT add-on (in case of AVX-512-related CPU dispatch issues) — no change
  9. Disabled two flaky Frigate camera streams (in case of a decode-related memory corruption path) — no change
  10. Switched back to CPU Type host — longest clean stretch yet (~9 hours overnight), but crashed twice again by morning

Checked the Proxmox host journal for hardware-level evidence: no MCE, no thermal throttling, one ambiguous EDAC/IBECC “memory error” log entry at first boot (matches a known igen6_edac false-positive pattern, not repeated since), and live corrected/uncorrected error counters read 0.

Where I’m at

Every layer I can influence through software, firmware, or config has been tested and ruled out — CPU virtualization settings, RAM, storage, container state, OS install, BIOS/microcode, and even application-level add-ons/cameras that touch CPU-intensive code paths. The crash signature is identical regardless of what changes. At this point it looks like a NUC13 board or CPU-level hardware issue, and I’m in the process of moving Home Assistant to older, non-hybrid-core hardware as a workaround while I pursue this with ASUS support.

Question for the community: has anyone seen SIGILL crashes like this on a hybrid P-core/E-core NUC13 (or similar 13th/14th gen Intel) running HA under Proxmox/KVM? Particularly interested if anyone has hit something similar? I don’t know where to go from here. In the past my HA installs have been rock solid with weeks of uptime.

Why Proxmox Ver 8, when you now create a new install from scratch?
And can you get any logs from the crash, to see whats crashes, what app, image, container …

You can get SIGILL if you download a docker image thats not made for x86, like aarch64.
(Try’s to run ARM ISA in X86 ISA env).

And when i asking the clanker, it says that maybe you running a python3 app that requere AVX512, and your CPU don’t have that it only support AVX and AVX2.

So if you can figure out witch python3 app that runs and SIGILL, you maybe can switch that to a compiled version with just AVX and AVX2.

Have you tried deploying the vm using the community script? Proxmox VE Scripts

What about adding an issue to core if you are certain it’s HA instead of having an AI write all this stuff up for you and posting it here?

¯\(ツ)/¯ Sounds like you are looking for…

Thank you. I will update Proxmox. I have not done that. I did try to identify if a python3 app, tried shutting down any add-on that may be requiring AVX but so far no luck. I reinstalled HAOS but that did not help either.

Thank you for the suggestion. I will try that path and see if it helps.

Thanks, still trying to figure out if this is something specific to me or a general issue. So far I have not heard that others are having similar issues.

Have you tried running a bare HA(OS), without restoring backups, just to see if the issue is with core or with one of the integrations?

After some further investigation and thanks to the comment to look further into the Home Assistant Core issues thread on Github. This seems to be a 2026.7.2 issue. Upgrading to 2026.07.3 did not fix the issue.
I did downgrade to 2026.6.4 which has been stable now for 7 hours.
Thank you for the suggestions.

Fix is merged. Looks like it’ll be fixed in the next release, either 2026.7.4 or 2026.8.0 ONVIF integration causes Fatal Python error: Illegal instruction on Alder Lake-N CPUs (i3-N305/N100) and crashes HA due to AVX requirement · Issue #176427 · home-assistant/core · GitHub