Some Silly Benchmark Results
One 20 April 1998, I released a new
version of the brem benchmark. The older results
(a lot of them!) are archived under this link.
I do not expect the results to be different by more than a percent or two,
so these are still useful.
What follows corresponds to results for the new version of brem.
results from my machine
Configuration: Pentium Pro 200 MHz (Gateway G6-200), 64 MB RAM, 256k cache,
Linux 2.0.32 (RedHat 5.0).
Absoft F77 (v4.4) optimized "-O" --- 195 seconds
Absoft F90 (v1.01) optimized "-O" --- 156 seconds(!)
Absoft F90, optimized "-O", on F77 source --- 207 seconds
f2c/gcc (driver: fort77), optimized "-O4" --- 267 seconds
Very very interesting that the F90 version of the code is substantially
faster! Using the F90 compiler to compile the stock F77 source yields suboptimal
results, but rewriting to use F90 idioms pays off.
Caveats ... the results for Pentium-class machines depend on alignment
of the variables in memory, and this can change the execution time by as
much as 30 percent! With the old Absoft compiler (v3.4), for some reason
the alignment is better when running in an xterm than when running under
a linux virtual console; 170 sec is typical for under X, while 220 sec
is typical when running under console. Proper alignment can be forced by
using the "-s" flag (static allocation) for Absoft F77. For more information,
Alignment Info Page.
modified: Wed May 6 09:03:41 EDT 1998