RunPod•3mo ago

SOS pod gpu errors

pod_id=usg9djjhmpjfpd

# nvidia-smi
Sat Sep 21 14:48:08 2024
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.161.07             Driver Version: 535.161.07   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce RTX 4090        On  | 00000000:A1:00.0 Off |                  Off |
|ERR!   36C    P0              48W / 450W |  24205MiB / 24564MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
+---------------------------------------------------------------------------------------+

# nvidia-smi
Sat Sep 21 14:48:08 2024
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.161.07             Driver Version: 535.161.07   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce RTX 4090        On  | 00000000:A1:00.0 Off |                  Off |
|ERR!   36C    P0              48W / 450W |  24205MiB / 24564MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
+---------------------------------------------------------------------------------------+

As you see, gpu is completely dead. We facing multiple errors like this. We faced it multiple times. It started since 19 september.

5 Replies

yhlong00000•3mo ago

I don’t see any errors on our end, but your memory usage is nearly full (24,205 MiB / 24,564 MiB). Could this be the issue? You might want to try a 48GB GPU

nevermindOP•3mo ago

Whats an "ERR!" than? We actually use multiple instances of this configuration and sometimes pods just "die" during its lifetime (it affects our flow control).

yhlong00000•3mo ago

I checked the error logs on the server and didn’t find anything significant. The ERR! in nvidia-smi usually indicates it’s unable to monitor or report certain metrics, which could be due to a hardware issue or a temporary glitch. However, if you’re encountering this frequently, it’s unlikely that all of our GPUs have the same issue. It might be worth trying a higher-end GPU to see if the problem persists. I suspect the high memory usage could be a contributing factor.

nevermindOP•3mo ago

Alright, if we face the same issue again - I'll report to this thread and try my best to lock the pod

yhlong00000•3mo ago

surething, feel free to share podId and timestamp here.

Gaming

Programming

SOS pod gpu errors