BENCHMARKS

Description

• I have benchmarked the CFHHM program, which is well vectorizable, on different vector and superscalar platforms. I get rather good proportionality to the theoretical MFLOPS ratings on both types of machines. This is an extremely valuable result because it allows high-confidence extrapolations to other processors.
• There are some additional benchmarks shown, including matrix multiplication using different loop orders, and very large array sizes. This tests more than only the raw processor speed.

Benchmark results by R. Krivec

• Matrix multiplication, including large arrays
• Matrix multiplication on PC (Linux)
• Matrix multiplication on workstations
• CFHHM benchmark
• CFHHM optimization report (recent optimizations)
• Various benchmarks (Postscript) including Vax, Convex, HP, SGI, etc.

Selected data from benchmarks

• Floating point efficiency A comparison of SPEC and Linpack numbers for some processors, compiled by R. K.

Articles, reports

• Lokalni benchmarki Clanek proti proglasanju neoptimiziranih privatnih programov za benchmarke.
• Parallelization of the CFHHM package (general, developed on the SPP-1000)

Benchmark links

• Benchmark links (BenchWeb, etc.).

Benchmark packages

• Matrix multiplication package used in my benchmarks (gzipped tar).
• CFHHM benchmark package is available (uses older version of programs; gzipped tar).

R. Krivec