RTX PRO 6000 Blackwell no longer usable — repeated 0xbadf3200 register-read errors

Graphics / Linux Linux
1

I was referred here by NVIDIA Customer Care for a Linux-side assessment before proceeding with a warranty claim.

System:

  • Ubuntu, kernel 6.17.0-29-generic
  • 2x NVIDIA RTX PRO 6000 Blackwell Workstation Edition
  • PCI devices: 0000:02:00.0 and 0000:04:00.0
  • 0000:02:00.0 remains fully operational
  • 0000:04:00.0 is still visible in lspci, but is not usable through the NVIDIA driver/CUDA

Relevant current state:

0000:02:00.0
Kernel driver in use: nvidia

0000:04:00.0
No "Kernel driver in use" entry

Only /dev/nvidia0 exists; /dev/nvidia1 is missing. Only 0000:02:00.0 appears under /proc/driver/nvidia/gpus/.

The kernel reports:

NVRM: The NVIDIA probe routine was not called for 1 device(s).

Retained kernel logs from the failure event also show repeated low-level errors:

NVRM: GPU0 gpuHandleSanityCheckRegReadError_GH100:
Possible bad register read: addr: 0xb830b0,
regvalue: 0xbadf3200,
error code: Unknown SYS_PRI_ERROR_CODE

This error repeated continuously during the event.

Current PCIe AER counters for both GPUs show:

TOTAL_ERR_COR 0
TOTAL_ERR_NONFATAL 0
TOTAL_ERR_FATAL 0

The affected GPU is therefore still visible on PCIe, but is no longer being operated by the NVIDIA driver, while the second identical GPU remains functional.

Could an NVIDIA moderator/engineer please advise whether the repeated 0xb830b0 / 0xbadf3200 register-read failures together with the loss of NVIDIA driver/CUDA availability are consistent with a hardware failure, and whether this GPU should proceed to warranty/RMA service?