Mysterious new CUDA issue hindering my work greatly

Aurora
1

Hey all,
I’m new here. I recently posted a GitHub repo that runs mining software on Aurora (to be expanded to other Ublue variants) using Toolbox.

It stopped working this morning with the error:

CUDA error: file too short

I dug around and found this:

ls -l /usr/lib64/libcuda.so.570.86.16
-rw-r--r--. 1 root root 0 Feb 17 22:52 /usr/lib64/libcuda.so.570.86.16

For some reason it has a 0-byte file size.

Here is my system info if it helps:

  • I am running Aurora-DX 41
  • Quadro RTX 5000
  • Driver Version: 570.86.16
  • CUDA Version: 12.8

It seemed to happen after the latest update, but the issue persisted when I rolled back to the pre-update image.

I am not sure if this is the proper place to post this. My time is extremely limited currently and this was my quickest option.

Thank you. I love Aurora and Ublue.

2

This is what what I see for that file in Bluefin-DX (which should be same):

❯ ls -l /usr/lib/libcuda.so.570.86.16 
-rwxr-xr-x. 3 root root 25609636 Jan  1  1970 /usr/lib/libcuda.so.570.86.16

~ 
❯ sha256sum /usr/lib/libcuda.so.570.86.16
0534de781601dec878f51f7478e38d58664b42e1ce4bdf2996cf7384cbc40372  /usr/lib/libcuda.so.570.86.16

Clearly your system’s atomic partition is broken. Did any update fail? Can you boot into the previous installation?

I would suggest re-paving the OCI image to see if it works.

3

Thank you, re-paving worked. I am new to Atomic distros so it wasn’t clear to me that the atomic partition was broken.

I hope my initial post didn’t come off as rude. My social skills are not the best. I wasn’t complaining and apologize if it came off that way.

Much appreciated.