duncan@super.ORG (Duncan A. Buell) (06/29/90)
> From: brooking@mcnc.org (Jim Brooking) > What's a MIPS doing in c.s.s? We talk MEGAFLOPS here, boy! > They talk MIPS in alt.evil or other kinky newsgroups. > Jim Brooking........North Carolina Supercomputing Center.......(919)248-1145 Since I'm sure that this comment was made (at least in part) (I hope) in jest, I will respond in kind. I am a realist, and a second or third class citizen in the world of computing, and I understand that much (although not all by any means) historical experience might well allow one to conclude (as some have) (although erroneously) that "supercomputing" and "scientific supercomputing" are synonyms. (I have even seen this error promulgated in books on the subject, but after all, it is rare that books get published without some typos and errors.) But to return to the point: as we all know, history is a record of what happened and not necessarily a record of what should have happened. So let us go forward to the future without carrying with us the prejudices and misconceptions of the past. [Remove tongue from cheek.] *One* benchmark of computing power and speed is mflops, whether expressed as machoflops or mflops actually realized by real programs getting real answers. This benchmark, however, is only valid if the computation being benchmarked uses floating point or needs to, and not all large scale computations are of this type. For my part, I prefer a rating which *does* look more like a mips rating. This is, admittedly, because I don't have much use for floating point in my philosophy and because machines which emphasize floating point tend to do fixed point arithmetic and other things badly if at all. Rating a Vax 780 as a 1, then, our Convex C1 comes in at about 5 and our Cray 2 varies from a 3 to a 25. This is, of course, only speed, and does not take into account the big memory on the Cray; if memory is a necessity, sometimes one has to go to the Cray even if it's slower than a Sun workstation. I suggest we leave the floor open for discussing *all* kinds of large scale computing, ok? That way all of us who have consumed 22 Cray-2 cpu-months will feel welcome. Duncan Buell Supercomputing Research Center 301-805-7350 17100 Science Drive duncan@super.org Bowie, Maryland 20715