Unexpected NULL returns when reading GPU frame buffer memory regions on L4 instance

Google Cloud Compute Infrastructure
,
1

Hi Google Cloud Support,

I’m running a computational workload on an L4 GPU instance (project: [PII Removed by Staff]) and I’m observing unexpected behavior when attempting to read certain regions of the GPU frame buffer memory.

Specifically, when I use standard CUDA memory access patterns to read from the GPU frame buffer, certain memory regions consistently return NULL values — not errors, not uninitialized data, but NULL. This occurs regardless of privilege level. I’ve tested with both user-space and root-level access.

The GPU is otherwise functioning normally — nvidia-smi reports the device correctly (NVIDIA L4, Driver 570.211.01, CUDA 12.8), CUDA toolkit operations work, and my rendering workload executes without errors.

I’m trying to understand whether this NULL behavior in certain frame buffer regions is expected, a driver issue, or a hardware anomaly. I’ve not been able to find documentation explaining why portions of the frame buffer would be non-readable.

Instance details:

  • Project: (PII Removed by Staff)
  • Instance: layer-zero-proof
  • Zone: us-central1-a
  • GPU: NVIDIA L4
  • Driver: 570.211.01
  • CUDA: 12.8
  • OS: Ubuntu 22.04

Any guidance would be appreciated. Happy to provide additional logs or run diagnostic commands.

Looking for an Update!… let me know if I need to pay for support

Thank you,
JP

1 Like
2

can anyone let me know if this case is being worked at all