proteneer / timemachine

Differentiate all the things!
Other
138 stars 17 forks source link

Removes Neighborlist sync from Nonbonded potentials #1371

Closed badisa closed 1 week ago

badisa commented 2 weeks ago

Benchmarks

A10 Cuda Arch 8.6

Lose some performance on the non-RBFE simulations (no ixn-group, pairlist) due to the overhead of running no-op kernels but make up for it in the RBFE simulations and local MD. About 5% for RBFE and 8% for local MD

Master

dhfr-apo: N=23558 speed: 698.28ns/day dt: 2.5fs (ran 100000 steps in 30.94s)
dhfr-apo-barostat-interval-25: N=23558 speed: 630.02ns/day dt: 2.5fs (ran 100000 steps in 34.29s)

hif2a-apo: N=8805 speed: 1213.16ns/day dt: 2.5fs (ran 100000 steps in 17.81s)
hif2a-apo-barostat-interval-25: N=8805 speed: 1036.86ns/day dt: 2.5fs (ran 100000 steps in 20.84s)
hif2a-rbfe-barostat-interval-25: N=8840 speed: 868.29ns/day dt: 2.5fs (ran 100000 steps in 24.88s)
hif2a-rbfe-local: N=8840 speed: 1280.69ns/day dt: 2.5fs (ran 100000 steps in 16.87s)
hif2a-rbfe-barostat-interval-25-water-sampling-interval-400: N=8840 speed: 802.98ns/day dt: 2.5fs (ran 100000 steps in 26.90s)

solvent-apo: N=6282 speed: 1690.23ns/day dt: 2.5fs (ran 100000 steps in 12.78s)
solvent-apo-barostat-interval-25: N=6282 speed: 1347.52ns/day dt: 2.5fs (ran 100000 steps in 16.03s)
solvent-rbfe-barostat-interval-25: N=6317 speed: 1086.41ns/day dt: 2.5fs (ran 100000 steps in 19.89s)
solvent-rbfe-local: N=6317 speed: 1377.21ns/day dt: 2.5fs (ran 100000 steps in 15.69s)

PR

dhfr-apo: N=23558 speed: 640.36ns/day dt: 2.5fs (ran 100000 steps in 33.73s)
dhfr-apo-barostat-interval-25: N=23558 speed: 583.22ns/day dt: 2.5fs (ran 100000 steps in 37.04s)

hif2a-apo: N=8805 speed: 1167.89ns/day dt: 2.5fs (ran 100000 steps in 18.50s)
hif2a-apo-barostat-interval-25: N=8805 speed: 1005.05ns/day dt: 2.5fs (ran 100000 steps in 21.49s)
hif2a-rbfe-barostat-interval-25: N=8840 speed: 916.01ns/day dt: 2.5fs (ran 100000 steps in 23.58s)
hif2a-rbfe-local: N=8840 speed: 1378.70ns/day dt: 2.5fs (ran 100000 steps in 15.67s)
hif2a-rbfe-barostat-interval-25-water-sampling-interval-400: N=8840 speed: 687.72ns/day dt: 2.5fs (ran 100000 steps in 31.41s)

solvent-apo: N=6282 speed: 1627.12ns/day dt: 2.5fs (ran 100000 steps in 13.28s)
solvent-apo-barostat-interval-25: N=6282 speed: 1312.01ns/day dt: 2.5fs (ran 100000 steps in 16.47s)
solvent-rbfe-barostat-interval-25: N=6317 speed: 1188.85ns/day dt: 2.5fs (ran 100000 steps in 18.17s)
solvent-rbfe-local: N=6317 speed: 1552.69ns/day dt: 2.5fs (ran 100000 steps in 13.92s)
badisa commented 1 week ago

Investigating why hif2a-rbfe-barostat-interval-25-water-sampling-interval-400 slows down by 15% this change.