Forgot your password?
Document Actions
Cray C90
| Cray UNICOS machines have a hardware
performance monitor (hpm), which gives the number of floating point operations
per CPU second (FLOPS) performed by a given process. The FLOPS for other
machines are determined from C90 FLOPS and the ratios of the Zone-Cycles/sec. The FLOPS increases somewhat with problem size, e.g. by 13 percent in going from a 32-cubed problem to a 64-cubed problem. The FLOPS for a 128 x 64 x 64 problem is only about 2 percent greater than it is for a 64-cubed run. For ZEUS-MP, a 128-cubed problem cannot be run interactively due to memory restrictions, so I have used the FLOPS/(Zone-Cycles/sec) conversion factor from the 64-cubed run. Note that although ZEUS-MP uses algorithms conceptually the same as those in ZEUS-3D and practically every loop vectorizes, the C90 runs ZEUS-3D twice as fast as it runs the current version of ZEUS-MP. Remarkably, the CPU time spent in loops with strides greater than 1 is more than double the amount for a similar loop with stride 1 (the stride depends on the direction of the sweep in advection substeps). All routines were compiled with: cft77 -ez ZEUS-MP (10 steps)
GRID: 32 x 32 x 32 per processor (tile)
Processors tused(s) Zone-Cycles/sec MFLOPS C90s Speedup Speedup/Processor
1 2.69 120144 109.54 1.00 1.00 1.00
GRID: 64 x 64 x 64 per processor (tile) (10 steps)
Processors tused(s) Zone-Cycles/sec MFLOPS C90s Speedup Speedup/Processor
1 13.50 190408 150.96 1.00 1.00 1.00
GRID:128 x 64 x 64 per processor (tile) (10 steps)
Processors tused(s) Zone-Cycles/sec MFLOPS C90s Speedup Speedup/Processor
1 24.41 210263 163.65 1.00 1.00 1.00
ZEUS-3D (10 steps)
GRID: 32 x 32 x 32 per processor (tile)
Processors tused(s) Zone-Cycles/sec MFLOPS C90s Speedup Speedup/Processor
1 1.68 195137 184.42 1.00 1.00 1.00
GRID: 64 x 64 x 64 per processor (tile) (10 steps)
Processors tused(s) Zone-Cycles/sec MFLOPS C90s Speedup Speedup/Processor
1 7.70 340437 289.53 1.00 1.00 1.00
GRID:128 x 64 x 64 per processor (tile) (10 steps)
Processors tused(s) Zone-Cycles/sec MFLOPS C90s Speedup Speedup/Processor
1 12.52 418737 350.14 1.00 1.00 1.00
Back to Scaling Comparison Main |
The Computational Astrophysics Data Analysis Center