Comparison of timings for some production codes

MAIN CONCLUSIONS:

For Pentia (II), even for big memory problems, with irregular data, the times scale almost exactly with CPU speed.
The (now older) 500 MHz DEC Alpha works on my codes, with Red Hat Linux, at about the speed of a 300 MHz Pentium II.
Dual Pentium machines run nearly twice as fast --- only noticeable slow down was combinatorial optimization on large systems (27 M) --- about 10-20% slowdown.

Dual Pentium: 233 MHz, DELL, Fall '97, 64M RAM, Linux, IDE
Single Pentium: 300 MHz, DELL, Fall '97, Linux, 64M RAM, IDE
Alpha: 500 MHz, Aspen Systems, Fall '97, 128M RAM, Linux, SCSI
pop CPU: 266 MHz, Microtrends, Spring '98, 128M RAM, Linux 2.0.32, IDE
400 x 2: Dual 400 MHz (100 MHz bus of course), Dell, May '98, Linux 2.0.32, IDE
Codes below are generally integer intensive, barely use disk, and no careful tuning on the codes has been done. Note that though "arrays" are listed, the execution of these problems is not to step through the arrays - execution can take one on very irregular paths through memory.
However, note timings for the FFT problem.
For comparison, timings given by Intel for its configuration of a 300 MHz processor is SPECint95 of 11.7, while Digital has reported SPECint95 of 14.2 for its 500 MHz Alpha.

Problem Problem size Linux Alpha 500 Single Pentium 300 Dual Pentium 233- Elapsed Time for single job Dual Pentium 233 - Elapsed Time to complete two identical jobs Dual Pentium 266 (pop) - Elapsed time, single CPU Dual Pentium 266 (pop) - Elapsed time, both CPUs (two identical jobs) Dual Pentium 400 (pop) - Elapsed time, single CPU Dual Pentium 400 (pop) - Elapsed time, both CPUs (two identical jobs)

Simple Monte Carlo - random arrival times of photons (used in teaching) - 6.6 s 4.63 s - 6.1 s 5.14 s 5.22 s 3.53 s 3.54 s

Finding threshold for conduction in array of quantum dots - invasion percolation type of problem (regular array of floats, search problem) 100 x 100 1.7 s 1.9 s 2.5 s - 2.16 s 2.22 s 1.44 s 1.44 s

200 x 200 16.2 s 16.0 s 20.6 s 21.0 s 18.0 s 18.0 s 11.9 s 12.0 s

400 x 400 (5.5 M) 118.9 s 129.1 s 166.6 s - 146.7 s 145.3 s 96.4 s 96.7 s

Combinatorial Optimization using integers on a regular lattice 10 samples of 480 x 480 (8.5 M) 56.9 s 51.9 s 64.1 s 68.9 s 57.5 s 55.8 s 33.8 s 36.7 s

2 samples of 720 x 720 (27 M) 24.0 s 26.1 s 30.5 s 37.9 s 25.1 s 27.8 s 16.7 s 18.2 s

Fast Fourier Transform, 2 dimensions 200 x 200, 50 samples 17.1 14.8 14.9 10.2 10.2

800 x 800, 10 samples (11 M) 67.3 60.0 59.9 44.2 44.3

1600 x 1600, 10 samples (41 M) 296 296 299

Geometric Mean Intel MHz per CPU, normalized to Single 300 291 300 238 273 269 407 420

Email: aam@syr.edu

Problem	Problem size	Linux Alpha 500	Single Pentium 300	Dual Pentium 233- Elapsed Time for single job	Dual Pentium 233 - Elapsed Time to complete two identical jobs	Dual Pentium 266 (pop) - Elapsed time, single CPU	Dual Pentium 266 (pop) - Elapsed time, both CPUs (two identical jobs)	Dual Pentium 400 (pop) - Elapsed time, single CPU	Dual Pentium 400 (pop) - Elapsed time, both CPUs (two identical jobs)
Simple Monte Carlo - random arrival times of photons (used in teaching)	-	6.6 s	4.63 s	-	6.1 s	5.14 s	5.22 s	3.53 s	3.54 s
Finding threshold for conduction in array of quantum dots - invasion percolation type of problem (regular array of floats, search problem)	100 x 100	1.7 s	1.9 s	2.5 s	-	2.16 s	2.22 s	1.44 s	1.44 s
	200 x 200	16.2 s	16.0 s	20.6 s	21.0 s	18.0 s	18.0 s	11.9 s	12.0 s
	400 x 400 (5.5 M)	118.9 s	129.1 s	166.6 s	-	146.7 s	145.3 s	96.4 s	96.7 s
Combinatorial Optimization using integers on a regular lattice	10 samples of 480 x 480 (8.5 M)	56.9 s	51.9 s	64.1 s	68.9 s	57.5 s	55.8 s	33.8 s	36.7 s
	2 samples of 720 x 720 (27 M)	24.0 s	26.1 s	30.5 s	37.9 s	25.1 s	27.8 s	16.7 s	18.2 s
Fast Fourier Transform, 2 dimensions	200 x 200, 50 samples	17.1				14.8	14.9	10.2	10.2
	800 x 800, 10 samples (11 M)	67.3				60.0	59.9	44.2	44.3
	1600 x 1600, 10 samples (41 M)	296				296	299
Geometric Mean Intel MHz per CPU, normalized to Single 300		291	300	238		273	269	407	420