I have a small cluster of 4 nodes, and each node contains 2 Tesla k40m cards. We’ve been using the HPL specific for FERMI that is off of the Nvidia Developer Zone website.
Earlier in this board, there is mention of modifying the Fermi version of HPL to make a more generic HPL for cuda:
Does anyone know any specifics on how I can disable the fermi portion of the makefile?
It appears that the makefile requires all the source files in the src/cuda folder to build libdgemm, so it is unclear to me what I need to remove in order to enable a broader cuda version.