skip to content
 

The CUDA Toolkit is a C language development environment for CUDA-enabled GPUs.

Availability: 

All 64-bit managed Linux workstations but only useful on machines with high-end NVIDIA graphics cards and with the CUDA driver installed. The CUDA driver is not installed by default.

Instructions for users: 

You need to load the 'cuda' module. On the workstations the NVIDIA CUDA software developer kit (SDK) can be found on the NFS server at the same place as the cuda installation. usually this is: /usr/local/shared/ubuntu-$(lsb_release -rs)/x86_64/cuda/ . The installer for the CUDA driver is usually also in that directory.

Licence Details: 
Documentation: 
Admin notes: 

You need to match the version of CUDA, the NVIDIA driver, and the SDK on each machine. Also there are different versions of CUDA for different Linux releases,


Once you've installed CUDA and the SDK you need to build libcutil.a (a utility library that the SDK examples reply on) by going to the common subdirectory and running make. Then you can build things out of the projects directory. If you've got the NFS server mounted readonly you can set the variables BINDIR and ROOTOBJDIR to places you can write. You need the freeglut-devel package installed to build some of the examples.


The SDK won't install unless you already installed CUDA and have it in your PATH. The SDK will also sometimes appear to hang while installing. It's waiting for you to type the path to install in at it.


A simple test of a CUDA install is to build the deviceQuery example and run it.

Can't find what you're looking for?

Then you might find our A-Z site index useful. Or, you can search the site using the box at the top of the page, or by clicking here.