UF Research Computing has added 104 GPUs to the cluster. These GPUs will be available to those who have made GPU investments.
The new GPU hosts will be configured with device drivers supporting the CUDA 10 Toolkit and CUDA 10 is the intended host platform for the new GPUs. Applications compiled with CUDA 9 may need to be recompiled for the newer CUDA toolkit. The popular applications installed on the cluster are being upgraded at this time.
Since some CUDA 9 applications may not work properly under the CUDA 10 installation, some GPU nodes will be kept at the CUDA 9 environment until October 31, 2019 to allow users to continue using their applications. Users are encouraged to convert to CUDA 10 now.
If your application needs a CUDA 9 environment you can request the cuda9 node feature.
For more information on using these cards please refer to the Research Computing Wiki page for GPU Access.
If you need help in compiling your application for CUDA 10 or have other problems, please file a support request.