Community Atmosphere Model (CAM) Performance
Scaling
Performance on all machines
| Platform | Tasks | Threads | Wallclock Time (sec)/simulation-day |
| IBM-SP (Hybrid) | 2 | 4 | 196.6 |
| IBM-SP (Hybrid) | 4 | 4 | 111.5 |
| IBM-SP (Hybrid) | 8 | 4 | 64.9 |
| IBM-SP (Hybrid) | 16 | 4 | 39.8 |
| IBM-SP (Pure MPI) | 2 | 1 | 594.5 |
| IBM-SP (Pure MPI) | 4 | 1 | 337.0 |
| IBM-SP (Pure MPI) | 8 | 1 | 187.8 |
| IBM-SP (Pure MPI) | 16 | 1 | 108.9 |
| IBM-SP (Pure MPI) | 32 | 1 | 67.7 |
| IBM-SP (Pure MPI) | 64 | 1 | 48.9 |
| SGI-O3 (Pure Open-MP) | 1 | 16 | 85.3 |
| SGI-O3 (Pure Open-MP) | 1 | 32 | 57.2 |
| SGI-O3 (Pure Open-MP) | 1 | 64 | 49.9 |
| SGI-O3 (Pure MPI) | 16 | 1 | 80.8 |
| SGI-O3 (Pure MPI) | 32 | 1 | 47.9 |
| SGI-O3 (Pure MPI) | 64 | 1 | 31.0 |
| Compaq-Alpha (Hybrid) | 2 | 4 | 112.8 |
| Compaq-Alpha (Hybrid) | 4 | 4 | 64.7 |
| Compaq-Alpha (Hybrid) | 8 | 4 | 37.4 |
| Compaq-Alpha (Pure MPI) | 4 | 1 | 181.3 |
| Compaq-Alpha (Pure MPI) | 8 | 1 | 107.5 |
| Compaq-Alpha (Pure MPI) | 16 | 1 | 62.3 |
| Compaq-Alpha (Pure MPI) | 32 | 1 | 38.8 |
| Linux-Lahey (Pure MPI) | 2 | 1 | 714.4 |
| Linux-PGI(Pure Open-MP) | 1 | 2 | 559.3 |
| Solaris (Pure MPI) | 2 | 1 | 1473.6 |
All tests were performed with CAM2.0.dev12 in August, 2002.
The time-metric is the wall-clock time for "stepon" for a standard T42 Eulerian-dynamics 10 day simulation, according to the standard timing library diagnostic that comes as part of CAM2.
Tests where the number of threads is one, were performed with Open-MP turned off. Likewise, tests where the number of tasks is one, were performed with SPMD (MPI) turned off.
- Compaq-Alpha results were performed on colt.ornl.gov (@ Oak Ridge National Lab) (OSF1 V5.1 732, f90 X5.4A-1684, 16 nodes, With four 667 MHz Alpha EV67 processors and 2GB RAM per node)
- IBM-SP results were performed on blackforest.ucar.edu (AIX 4.3.3.79, F90 7.1.1.2, 293 WinterHawk II nodes with four 64-bit 375-MHz POWER3 CPUs and 2 GB of memory per node.)
- Linux-Lahey results were performed on apache.cgd.ucar.edu with the Lahey F95 compiler (Lahey lf95 L6.10a, 2-cpu Intel Pentium-III 1133MHz, 770MB RAM)
- Linux-PGI results were performed on apache.cgd.ucar.edu with the Portland-group F90 compiler (pgf90 3.3-2, 2-cpu Intel Pentium-III 1133MHz, 770MB RAM)
- SGI-O3 results were performed on chinook.ucar.edu (MIPSpro 7.30, 128 500-MHz R14000 CPUs, 64 GB distributed shared memory )
- Solaris results were performed on sanitas.cgd.ucar.edu (Sun WorkShop 6 update 2, Solaris 5.6, 2-CPU ultra-2, 296MHz processor, with 769 MB RAM)
CESM Project
The Community Earth System Model (CESM) is a fully-coupled, global climate model that provides state-of-the-art computer simulations of the Earth's past, present, and future climate states.
CESM is sponsored by the National Science Foundation (NSF) and the U.S. Department of Energy (DOE). Administration of the CESM is maintained by the Climate and Global Dynamics Laboratory (CGD) at the National Center for Atmospheric Research (NCAR).