How to get a specific version of cuda?

Hi, I use the Tensorflow 1.15.4 which relies on cuda 10.0 and cudnn 7.4.2. These libs are available on previous HPC but missing for the discovery cluster. How can I install these versions of libs on the discovery cluster? Or can you help me installing it?

Thanks

module spider cuda


cuda:

 Versions:
    cuda/8.0.61
    cuda/9.0.176
    cuda/10.1.243
    cuda/10.2.89
    cuda/11.0.2

@xusenyin I’ve installed cuda/10.0.130 and a corresponding cudnn/7.6.5.32-10.0-cuda module. They will be available on the cluster in an hour or so. That cudnn version is compatible with cuda 10.0, so it should work in place of the older version you referenced. Let me know if you encounter any issues though.

@dstrong Thanks, it works great as I load them directly with the names

module load cuda/10.0.130
module load cudnn/7.6.5.32-10.0-cuda

But when I use module spider cuda, the cuda/10.0.130 is not shown in the list.

It seems the module spider cache has not yet updated to find the new modules. You can use module --ignore-cache spider cuda instead.