This adds simple instruments and check_speed.py configurations used to check the performance of the HMS hollow cylinder and UN against the CPU version of mcvine.
@ckendrick I made some minor changes and added some timing results. Please check.
Some observations
On inception (EPYC 7662 vs 3080), the speedup for the UN multiple scattering case (ncount=1e9) is about 100X
On my 7 year old desktop (i7 4790K vs GTX 1660), it is about 36X
I was surprised that for this simulation the performance of individual core of the relative old i7 4790K is actually slightly better than individual core of EPYC 7662. But considering 4790K is 4GHz 7662 is 2GHz it may not be wrong
For ncount=1e9 RTX3080 is only a bit more than twice as fast as GTX1660. Should run 1e10 and check
This adds simple instruments and
check_speed.py
configurations used to check the performance of the HMS hollow cylinder and UN against the CPU version of mcvine.