Impossible installation? CUDA5 for Tesla M10 card on Ubuntu 1604 ?


I wonder if it is now impossible to install CUDA5 for a Tesla M10 card on Ubuntu 1604.

This is a server which was donated recently which I would like to use.

I have tried:

  • Removing all previous CUDA (Cuda 9, Cuda 10) with apt-purge.
  • Trying to use the runfile sudo ./
  • Stopping the xserver - sudo service lightdm stop

I am now seeing a “kernel sources missing” error.

ERROR: The kernel header file
       '/lib/modules/4.15.0-47-generic/build/include/linux/version.h' does
       not exist.  The most likely reason for this is that the kernel
       source files in '/lib/modules/4.15.0-47-generic/build' have not been

Is this even remotely possible or am I barking up the wrong tree ?

Might it be a better strategy to try to install CUDA via nvidia-docker, or does this not extend way back to CUDA5 ?

Thanks for your attention!

CUDA 5 and any drivers associated with CUDA5 wouldn’t work with Tesla M10. That is a Maxwell product. CUDA 5 supported up through Fermi. Furthermore you have a relatively new kernel (4.15) and there aren’t any drivers that old (CUDA5 era) that would be installable on that kernel, most likely.

If you install a recent driver (such as that from the CUDA 9 or 10 toolkit installer) and can get that working with the M10, you could then try to install just the CUDA 5 toolkit. In theory that should work. Not sure why you would want to install such an old toolkit.

Thanks for your helpful reply Robert. I’ll give that a go.