[comp.sys.next] Comparing benchmarks

ms6b+@andrew.cmu.edu (Marvin Sirbu) (11/17/90)

I must thank Ralph Seguin for taking the time to enter benchmark results
under different test suites for a range of machines.

The most striking thing to me was the variation in relative performance of
machines depending upon the benchmark.  For example, taking two new
machines, the RIOS 320 (IBM RISC) and the Decstation 5000 (MIPS RISC):
on the three benchmarks the ratio of performance was 3.03, 1.8 and .66
respectively.  That is, on the Linpack benchmark the IBM looks three times
better than the Decstation, while on the Dhrystone benchmark, the Decstation
beats the IBM by 1.5 to 1.

Which just goes to prove the old adage:  don't rely on benchmarks--compare
machines on the kind of work you do!

Marvin Sirbu
Carnegie Mellon