NAMD Performance
NAMD 2.6 on ApoA1
92,224 atoms, 12A cutoff + PME every 4 steps, periodic
Time is in seconds per step, averaged over a number of steps, after load balancing. NAMD prints this data, labeled Benchmark time, twice near the beginning of a run. Due to its measurement-based load balancing system, NAMD performance may improve dramatically after the first couple of hundred timesteps.
Download this benchmark: config, directory, 2.8M archive
Previous versions: 2.5, 2.4, 2.3, 2.2
Independent Performance Analyses
Joachim Hein, Fiona Reid, Lorna Smith, Ian Bush, Martyn Guest, and Paul Sherwood. On the performance of molecular dynamics applications on current high-end systems. Philosophical Transactions: Mathematical, Physical and Engineering Sciences Volume 363, Number 1833, Pages 1987-1998, August 15, 2005. journal
AMBER group MD Benchmarks for Amber, CHARMM and NAMD (updated April, 2005)
DHFR Common CHARMM/AMBER Benchmark
23,558 atoms, 9A cutoff + PME every step, periodic, Origin 2000 R10000/250
New benchmarks on IBM platforms are available from the AMBER developers.
| NAMD 2.2 | NAMD 2.4b1 | NAMD 2.5b2 | AMBER 6 | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Procs | Time | Speedup | Time | Speedup | Time | Speedup | Time | Speedup | |||||
| 1 | 8.24 | 1 | 5.08 | 1 | 4.10 | 1 | MAX_RSTACK | ||||||
| 2 | 4.26 | 1.93 | 2.66 | 1.91 | 2.16 | 1.90 | 1.96 | 2 | |||||
| 4 | 2.19 | 3.76 | 1.35 | 3.76 | 1.11 | 3.69 | 1.05 | 3.73 | |||||
| 8 | 1.12 | 7.36 | 0.69 | 7.36 | 0.57 | 7.19 | 0.60 | 6.53 | |||||
| 16 | 0.57 | 14.5 | 0.35 | 14.5 | 0.29 | 14.1 | 0.38 | 10.3 | |||||
| 32 | 0.32 | 25.8 | 0.19 | 26.7 | 0.17 | 24.1 | 0.32 | 12.3 | |||||
| 64 | 0.20 | 41.2 | 0.11 | 46.2 | 0.10 | 41.0 | 0.25 | 15.7 | |||||
| 126 | 0.14 | 58.9 | 0.07 | 72.6 | 0.07 | 58.6 | 0.36 | 10.9 | |||||
| 252 | 0.09 | 91.6 | 0.06 | 84.7 | 0.07 | 58.6 | 0.76 | 5.16 | |||||
| config | config | config | config | ||||||||||
Time is in seconds per step, averaged over a number of steps, after load balancing. NAMD prints this data, labeled Benchmark time, twice near the beginning of a run. Due to its measurement-based load balancing system, NAMD performance may improve dramatically after the first couple of hundred timesteps.
The NCSA Origin was chosen for this benchmark because AMBER is installed and available on the machine. The slower processors (250 MHz) and fast interconnect on this machine should demonstrate best possible AMBER scaling.