Got a request to install VMD on our linux workstations and thought it would be rather straight forward. It is, but I found it out in a very round about way. To save you the long and dark road I took to get back in the sunshine, it went down like this:
These workstations have Nvidia graphics and I had most installed with rpm packages(dkms is used to rebuild the kernel module when new kernel is installed). I thought that was that but only to find the GLX module doesn’t load. Reading the forums I found that Xorg was still loading the original glx module, not the Nvidia one. Renamed the file(/usr/lib64/Xorg/modules/extensions/libglx.so), and glx now loads and so does the prebuilt VMD binaries.
BUT, yes a but. But turns out VMD is unable to detect the cards support CUDA extentions. Turns out for that, you have to go the old route of running the Nvidia installer, but that turns out to be a good thing as it fixes the glx mismatch issue. So while yes rpm is a nice solution for nvidia(with dkms), the module installer is the better way.