Convex Exemplar SPP-1600
| These runs were done by Greg Astfalk
of HP/Convex. Isom Crawford's Fortran-callable interface to the thread timing routines is available here. This 4-HYPERnode system was in single-user mode. All routines were compiled with: f77 -O; the code ran about 30 percent faster with f77 -O than with fc -O0.
GRID: 32 x 32 x 32 per processor(tile)
Speedup/
Processors Wall Clock tused(s) Zone-Cycles/sec MFLOPS C90s Speedup Processor
1 3.76 3.60 36034 32.85 0.30 1.00 1.00
2 3.83 3.63 71333 65.04 0.59 1.98 0.99
3 4.10 3.76 101439 92.49 0.84 2.82 0.94
4 4.18 3.76 137191 125.08 1.14 3.81 0.95
6 4.70 4.01 190218 173.43 1.58 5.28 0.88
8 5.31 3.98 260779 237.76 2.17 7.24 0.90
12 5.92 4.42 343548 313.23 2.86 9.53 0.79
16 6.19 4.65 434521 396.17 3.62 12.06 0.75
24 7.03 6.12 594503 542.03 4.95 16.50 0.69
32 6.63 5.28 771388 703.31 6.42 21.41 0.67
GRID: 64 x 64 x 64 per processor(tile)
Speedup/
Processors Wall Clock tused(s) Zone-Cycles/sec MFLOPS C90s Speedup Processor
1 29.94 29.01 35799 28.40 0.19 1.00 1.00
2 30.63 29.30 70900 56.25 0.37 1.98 0.99
3 32.35 30.58 101240 80.32 0.53 2.83 0.94
4 32.71 29.93 138851 110.16 0.73 3.88 0.97
6 36.90 32.20 191653 152.06 1.00 5.35 0.89
8 40.75 33.00 250997 199.14 1.31 7.01 0.88
16 48.56 39.09 408269 323.92 2.13 11.40 0.71
24 45.08 38.97 627165 497.59 3.28 17.52 0.73
32 51.49 42.80 771784 612.33 4.03 21.56 0.67
Back to Pure Hydro Main |