NERSC GAMESS Benchmark Performance Data
The first table lists processing rates (in GFLOPS/s) and corresponding machine efficiences. Floating-point operation counts were obtained using the NERSC IPM utility on Seaborg. Benchmark runtimes are presented in the second table. Machine names are hyperlinks to machine descriptions.
- The two problem sizes are not related in any way; they solve different chemical systems.
- Significantly different versions of the code have been used on different systems. For example, the version run on the Cray Jaguar XT3 and XT4 systems uses SHMEM for the medium problem and MPI for the large problem. The version run on the Cray Franklin XT4 system uses a mixed MPI/SHMEM method for both problem sizes.
All runs were conducted during normal production periods while a full workload was running. This can introduce a variability in the measurement that differs from machine to machine.
Detailed IPM Performance Counter Profiles
- GAMESS Medium: Bassi
- GAMESS Large: Bassi
|Machine||Processing Rates (in GFLOPS/s) and Machine Efficiencies|
|64 MPI Processes ("Medium" Case)||384 MPI Processes ("Large" Case)|
|Seaborg||3.9 (4.1%)||13.1 (2.3%)|
|Jacquard||20.4 (7.2%)||115.2 (6.8%)|
|Bassi||26.4 (5.4%)||120.8 (4.1%)|
|Jaguar XT3 Dual-Core||26.9 (8.1%)||156.4 (7.8%)|
|Jaguar XT4 Dual-Core||26.2 (7.9%)||158.4 (7.9%)|
|Franklin||22.6 (6.8%)||201.1 (10.1%)|
|Machine||Benchmark RunTime (seconds)|
|64 MPI Processes |
|384 MPI Processes
|Seaborg||18,665 s||42,055 s|
|Jacquard||3607 s||4767 s|
|Bassi||2788 s||4546 s|
|Jaguar XT3 Single-Core Small Pages|
|Jaguar XT3 Single-Core Large Pages||2634 s||3430 s|
|Jaguar XT3 Dual-Core Small Pages||-||-|
|Jaguar XT3 Dual-Core Large Pages||2736 s||3511 s|
|Jaguar XT4 Single-Core Small Pages||-||-|
|Jaguar XT4 Single-Core Large Pages||2700 s||3863 s|
|Jaguar XT4 Dual-Core Small Pages||-||-|
|Jaguar XT4 Dual-Core Large Pages||2806 s||3467 s|
|Franklin||3246 s||2731 s|