gillies@m.cs.uiuc.edu (07/16/90)
> So, the data and the claim that 68040==20VAXmips implies that the earlier > 68020 has a "sustained performance level of 4.9 VAX-equivalent MIPS" > (4.9 = 20/4.1). Does anybody seriously believe this? I thought that originally, a VAX-mip (= speed of Vax 11/780 on j random code) was .4 million instructions per second (.4 "native mips" ?). This would put the 68020 at 4.9*.4 ~ 2.0 "native mips". This is a widely quoted figure (2-3 mips) for the 68020, and believable for a 16Mhz system, judging from cycle counts in the motorola manual. On the other hand, a VAX-mip is an 11/780 mip, isn't it? A VUP is something entirely different, based on microvaxes, n'est-ce pas? disclaimer: This may be completely wrong, please don't flame me. Don W. Gillies, Dept. of Computer Science, University of Illinois 1304 W. Springfield, Urbana, Ill 61801 ARPA: gillies@cs.uiuc.edu UUCP: {uunet,harvard}!uiucdcs!gillies
mash@mips.COM (John Mashey) (07/30/90)
In article <3300154@m.cs.uiuc.edu> gillies@m.cs.uiuc.edu writes: > >> So, the data and the claim that 68040==20VAXmips implies that the earlier >> 68020 has a "sustained performance level of 4.9 VAX-equivalent MIPS" >> (4.9 = 20/4.1). Does anybody seriously believe this? > >I thought that originally, a VAX-mip (= speed of Vax 11/780 on j >random code) was .4 million instructions per second (.4 "native mips" >?). This would put the 68020 at 4.9*.4 ~ 2.0 "native mips". This is >a widely quoted figure (2-3 mips) for the 68020, and believable for a >16Mhz system, judging from cycle counts in the motorola manual. > >On the other hand, a VAX-mip is an 11/780 mip, isn't it? A VUP is >something entirely different, based on microvaxes, n'est-ce pas? > >disclaimer: This may be completely wrong, please don't flame me. 1) VUP: 1 VAX 11/780 == 1, on some (internal) set of about 100 benchmarks that DEC uses, mixture of languages, with fair amount of floating-point. As far as I know, DEC has generally given VAXen VUPs ratings, NOT mips- ratings. 2) MVUP = MicroVAX II Unit of Performance, used by Digital Review magazine, based on timings of about 30 FORTRAN codes. ------------ Regarding mips-ratings, and SPEC, and such: 1) Unadorned mips-ratings are essentially useless, and ought to be eradicated forever. Read the June 20, 1990 Wall Street Journal article ("MIPS Keeps Slipping as Speed Standard") for a succinct summary of the rationale for this statement. 2) Although vendors are (at least mostly) internally consistent, each has its own idea of mips-ratings, such that: if vendor A has a machine rated 10 mips and vendor B has one rated also at 10 mips it is quite possible for one of the two machines to run 1.75X faster on the SPEC integer subset. Put another way, SPEC integer performance ranges anywhere from 55% to 97% of the corresponding published mips-ratings, depending on vendor & product. (This was from 2Q90 data, may have changed by now.) 3) Any time you see, in the press, a statement like: "Vendor A announces new XX-mips machine", unless you have other data, or are very familiar with conversion factors from that product line to others, you would be wise to replace it with: "Vendor A announces machine, for which no meaningful performance rating has been given." -- -john mashey DISCLAIMER: <generic disclaimer, I speak for me only, etc> UUCP: mash@mips.com OR {ames,decwrl,prls,pyramid}!mips!mash DDD: 408-524-7015, 524-8253 or (main number) 408-720-1700 USPS: MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086