bdot - Memory/Cache/FPU Benchmark Jan Mandel, January 1996 http://www-math.cudenver.edu/~jmandel This tests mainly memory speed; the loop has one load from memory per one double precision multiply/add operation. It reveals cache sizes and systems with unbalanced CPU vs. memory throughput quite clearly. The main loop do i=1,n t=t+x(i)*x(i) enddo is admittedly not very useful but it is believed the CPU and memory utilization pattern is characteristic of iterative methods for large systems of equations. The results were edited by deleting uninteresting rows. ----------------------------------------------------------------- SPEED OF SUM X(I)*X(I), I=1,N SGI Power Challenge L 75MHz (dracula) f77 -O3 IRIX 6.1 N kB MB MFLOPS UNROL 2 10000 78.1 0.1 198.08 198.23 100000 781.2 0.8 198.22 199.34 500000 3906.2 3.8 156.70 183.89 <-- cache 4MB 1000000 7812.5 7.6 40.16 39.99 3000000 23437.5 22.9 33.62 33.54 IBM RS/6000 360 (tiger) xlf 3.1 f77 -O3 AIX 3.5 N kB MB MFLOPS UNROL 2 1000 7.8 .0 54.05 51.28 4000 31.2 .0 52.63 47.62 <-- cache 32kB 10000 78.1 .1 28.17 27.40 100000 781.2 .8 29.85 30.30 1000000 7812.5 7.6 28.17 28.99 2000000 15625.0 15.3 29.41 28.57 Sparcenter 1000E (math) f77 -fast -O4 Solaris 2.1 N kB MB MFLOPS UNROL 2 1000 7.8 0.0 30.76 30.48 5000 39.1 0.0 25.52 25.29 50000 390.6 0.4 25.41 24.70 100000 781.2 0.8 24.39 24.04 <-- cache 1MB 200000 1562.5 1.5 13.28 13.47 500000 3906.2 3.8 11.19 11.18 3000000 23437.5 22.9 11.23 10.91 DEC Alpha (carbon) f77 -fast -O4 (OSF 1) N kB MB MFLOPS UNROL 2 1000 7.8 0.0 133.93 134.81 <-- cache 8kB 2000 15.6 0.0 51.88 52.95 10000 78.1 0.1 50.47 50.85 20000 156.3 0.2 43.88 41.57 50000 390.6 0.4 42.51 45.04 100000 781.3 0.8 44.94 45.24 <-- cache 1MB 200000 1562.5 1.5 19.84 19.86 300000 2343.8 2.3 17.00 16.79 10000000 78125.0 76.3 16.81 16.77 IBM RS/6000 250 PowerPC 601 66MHz (putr) AIX 3.5 xlf 3.1 N kB MB MFLOPS UNROL 2 1000 7.8 .0 43.48 43.48 3000 23.4 .0 42.55 43.47 4000 31.2 .0 38.46 37.04 <-- cache 32kB 5000 39.1 .0 18.18 18.18 50000 390.6 .4 18.35 18.18 100000 781.2 .8 18.18 18.18 5000000 39062.5 38.1 17.86 18.02