Here's an excerpt from the final version of the Blitz++ ISCOPE'98
paper, showing benchmark results on a variety of platforms.
You can download the full paper from
http://oonumerics.org/blitz/papers/iscope98.ps
Table 1 shows performance of Blitz++ arrays on 21
loop kernels used by IBM for benchmarking the
RS/6000. Performance is reported as a fraction of Fortran
performance: > 100 is faster, and < 100 is slower.
The fastest native Fortran compiler was used, with typical
optimization switches (-O3, -Ofast).
The loop kernels and makefiles are available as part of the
Blitz++ distribution.
Platform/ Out of cache In-cache
Compiler Mean Median Mean Median
Cray T3E/KCC 95.7% 86.4% 98.1% 88.4%
HPC-160/KCC 100.2% 97.5% 95.1% 93.4%
Origin 2000/KCC 88.1% 87.3% 79.8% 78.6%
Pentium II/egcs 98.4% 98.5% 79.6% 82.6%
RS 6000/KCC 93.5% 90.7% 97.3% 93.2%
UltraSPARC/KCC 91.1% 86.8% 79.0% 78.3%
--------------------- blitz-dev list --------------------------------
* To subscribe/unsubscribe: mail to majordomo@oonumerics.org, with
"subscribe blitz-dev" or "unsubscribe blitz-dev" in the body of the message
* Blitz++ web page: http://oonumerics.org/blitz/
This archive was generated by hypermail 2b29 : Wed Feb 20 2002 - 04:30:06 EST