GROMACS version: 2020.4
GROMACS modification: No
Hi all, and happy new year! :)
I finally decided to give gpgpu a try in a notebook (i5 8300H + pascal GTX1050M), so I’m new to it. On the benchmark system I used (two proteins in water with PME, ~238,000 atoms) the performance jumped from 3.143ns/day to 7.899ns/day.
Since the results were interesting (no specific optimizations already led to 2.5x speedup on such a simple gpu), I began to wonder: How to predict scalability across different card models? For example, how much better would the RTX3060 and RTX3070 perform? What would be the (expected usual) optimal number of CPU threads for each card?
Thanks a lot in advance! :)