[cfarm-users] Request for tools on cfarm107/108

[Mosè Giordano]

Hello,

first of all thanks for cfarm107/108, they look very useful for
developing/testing GPU code!  I have some requests for these systems:

* install nvtop, to monitor GPU usage (nvidia-smi provides some of the
same information already, but nvtop is arguably nicer to look at)
* install Nsight Systems and/or Nsight Compute.  These are profiling
tools, which can also be installed in userspace so in this sense they
aren't critical, but I'd expect other users may want to use them and
so providing them with a single central installation would reduce the
duplication on disk
* related to the above point, allow all users, not just
administrators, to access performance counters when using the
profiling tools.  This requires setting the parameter
NVreg_RestrictProfilingToAdminUsers=0 for the nvidia kernel module,
see https://developer.nvidia.com/nvidia-development-tools-solutions-err_nvgpuctrperm-permission-issue-performance-counters.
I guess this could be the more contentious request, I'll leave the
decision of what to do to you of course.  Perhaps this could be
enabled on only one of the two machines, if deemed too risky.

Bye,
Mosè