jim@thrush.Stanford.EDU (Jim Helman) (05/27/89)
The May issue of IEEE CG&A has an article on benchmarking graphics workstations. From my limited knowledge of performance numbers, the he graphics benchmark results seem reasonable. However, the stated results from the CPU benchmarks indicate that the SGI 4D/60 (MIPS R2000/R2010 8MHz) is SLOWER than the old SGI 2400T (68020 16MHz) and LESS than 1/4 the speed of a Sun-3/260 (68020 25MhZ)! From a graph comparing CPU performance in the article: Est. from graph Reported Computer % of Vax8600 Seconds Vax 8600 100% 30 Sun 3/260 100% 30 Apollo DN580 30% 99 SGI 2400T 30% 64 (inconsistent, gives 46%) Iris 4D/60 25% 118 Vax 11/750 VMS 20% 149 Vax 11/750 Unix 5% 489 The benchmark was a multivariate least-squares program. Although the article doesn't mention what language, compiler optimization level, compiler version, level of FP precision, or the floating point hardware that were used, the 4D/60 results seem very unlikely to me regardless of what combination of these was used. I've benchmarked a variety FP intensive C code (including applications using both 32bit and 64bit FP). The 4D/70 (MIPS R2000/R2010 12.5MHz) was 3-6 time FASTER than a 3/260 w/FPA board. Even derating the 4D/60 for a slower clock speed of 8MHz, this is more than a factor of 10 off from their results. I don't think even RISC-bashers like Neal Nelson et al. could come up with a benchmark that could generate this sort of discrepency. As someone who has had to spend (spelled w-a-s-t-e) lots of time running benchmarks because some vendors' numbers can't be trusted, I think that independent benchmark reports like this article are invaluable. But it bothers the hell out of me when apparently misleading numbers and incomplete descriptions like these get published in a widely distributed publication. It's a good thing I know that my numbers are the only right ones :-), otherwise I might have acted on information like this. Jim Helman Department of Applied Physics P.O. Box 10494 Stanford University Stanford, CA 94309 (jim@thrush.stanford.edu) (415) 723-4940