Little program that makes use of HWLOC bindings into CUDA-driver/-runtime to find the nearest CPU (NUMA) to the GPU device(s).
Can be useful for configuring systems that bind to CPUs/Cores using some index value.
Requires CUDA 10+ and HWLOC v2, uses CMake to generate the build system:
$ mkdir build && cd build
$ cmake ..
$ make
$ ./nearest-gpu
Use to set the gres.conf
for each GPU device, example of what this looks
like:
TODO