Issue with CUDA 12 Update: cudaErrorDevicesUnavailable on Tesla M40 (Ubuntu Server)

Accelerated Computing CUDA CUDA Setup and Installation
1

Hi everyone,

I’m encountering an issue trying to upgrade to CUDA 12 on my server with 4 Tesla M40 24GB GPUs. Here are the details:

OS Ubuntu 24.04.1
GPUs: Tesla M40 24GB (4 units)
Working Setup: Driver Version: 470.256.02 CUDA Version: 11.4

trying to update to anything with cuda 12.x and drivers 530 and above results in the following situation:
nvidia-smi works - showing all gpus as expected, with the updated drivers/cuda version.
however, trying to run anything on the gpu results in a cudaErrorDevicesUnavailable error.

I can run deviceQuery, but not bandwidthTest (fails with cudaErrorDevicesUnavailable)

Any suggestions on how to resolve this?
Thanks in advance.

1 Like
2

Just a thought, if you used the pacjkages method of installing the driver - perhaps you’re using the open drivers, which aren’t compatible with older cards. I wouldn’t expect nvidia-smi to work in this case though.