walter@garth.UUCP (Walter Bays) (06/14/88)
Was: Maximum MIPS for a given memory bandwidth? >In article <22050@amdcad.AMD.COM> tim@amdcad.AMD.COM (Tim Olson) writes: >> Am29000 (Video DRAM) >>MIPS 12.7 Native, 15.2 VAX MIPS In article <491@daver.UUCP> daver@daver.UUCP (Dave Rand) writes: >I am confused. How can a risc machine have a higher "vax mips" than >native mips? MORE (not less) risc instructions are required to >do the same task, when compared to a vax. I've seen the same thing on many benchmarks of the Clipper: VAX MIPS are substantially higher than native MIPS. What the industry calls VAX MIPS is a speed ratio to a 780 (usually running some version of UNIX). The reason for the seeming anomaly is that our compilers are much more efficient than VAX UNIX compilers. If you translated the VAX assembler program to RISC assembler, more instructions would be required; but the pcc-derived compilers generate many more instructions to perform a given task than do most RISC compilers. VAX VMS (and, I've heard, Ultrix 2.0) compilers are much better. We often base our performance comparisons on VMS, but this has its own problems. First there's the nontrivial problem of porting a UNIX benchmark to VMS; the simpler it is the easier the port. Then there's the problem of the VMS optimizer eliminating large portions of the benchmark; the more complex it is the more meaningful the performance comparison. It's always a challenge to a benchmark author to accurately measure the benefit of an optimizing compiler on real applications. Benchmarks which are optimizable to "main(){}" grossly overestimate system performance. Benchmarks whose aim is to "defeat the optimizer" can grossly underestimate system performance. -- ------------------------------------------------------------------------------ My opinions are my own. Objects in mirror are closer than they appear. E-Mail route: ...!pyramid!garth!walter (415) 852-2384 USPS: Intergraph APD, 2400 Geng Road, Palo Alto, California 94303 ------------------------------------------------------------------------------