Running four molecular dynamics replicas at once¶
System(s): Polaris · Code: GROMACS · Outcome: Success
The ask¶
(This campaign predates verbatim prompt logging — the request below is reconstructed from the campaign's documented design, not a direct quote.)
"Run an ensemble of 4 independent molecular dynamics replicas simultaneously to test multi-replica throughput."
What happened¶
Rather than one simulation, Trinity launched 4 independent replicas of the same system concurrently on separate GPUs within a single job — a common technique for improving statistical sampling or exploring multiple starting conditions in parallel.
Results¶
- All 4 replicas ran concurrently and completed successfully, each on its own GPU.
- Per-replica throughput ranged from 1,479 to 1,617 nanoseconds/day — high enough that running replicas in parallel adds negligible overhead compared to running one alone.