[Nottingham] CPU and Memory benchmark programs

[Robert Davies]

Doing some experimentation comparing CPUs, compilers and optimisation 
options.  Surprisingly there seem to be few choices of CPU & Memory benchmark 
programs around, I've used one called nbench (byte benchmarks) and a program 
callled MemSpeed which shows nicely effect of cache sizes.

Any body got any suggestions for any better benchmark programs, apart from 
the obvious kernel compile?  Maybe I've missed the obvious?


Anyway FWIW, it would appear that the Gentoo optimised compile can bring 
decent speed ups on CPU bound stuff on i686 (and actually would expect Athlon 
and P4 to show greater effects though those systems should be comparatively 
more I/O bound than PII / Celeron ones).  I'm actually rather surprised by 
the improvement shown by gcc-3.2 over 2.95, and also the comparatively larger 
effect of -march= options in gcc-3.   The IDEA values were showing 50% 
improvement, which would suggest it's well worth the effort, for things like 
gzip / zlib, bzip2, ssh etc.  There's a significant improvment over 
-march=pentium options which was no better than portable i386 code on a i686 
core CPU.

C compiler          : gcc version 3.2
OPTFLAGS :  -O3 -fomit-frame-pointer
libc                : ld-2.2.5.so
INTEGER INDEX       : 7.692
FLOATING-POINT INDEX: 7.020
MEMORY INDEX        : 2.261
INTEGER INDEX       : 1.697
FLOATING-POINT INDEX: 3.893

C compiler          : gcc version 2.95.3 20010315 (SuSE)
OPTFLAGS :  -O3 -fomit-frame-pointer -arch=i686
libc                : ld-2.2.4.so
INTEGER INDEX       : 8.221
FLOATING-POINT INDEX: 6.712
MEMORY INDEX        : 1.992
INTEGER INDEX       : 2.097
FLOATING-POINT INDEX: 3.723

INTEGER INDEX       : 9.564
FLOATING-POINT INDEX: 7.084


C compiler          : gcc version 3.2
OPTFLAGS :  -O3 -fomit-frame-pointer -arch=pentium2
libc                : ld-2.2.5.so
INTEGER INDEX       : 9.564
FLOATING-POINT INDEX: 7.084
MEMORY INDEX        : 2.243
INTEGER INDEX       : 2.500
FLOATING-POINT INDEX: 3.929

Rob