ted@imsvax.UUCP (Ted Holden) (01/23/87)
Kenneth Ng of Argus writes: >By the way, here is an actual benchmark done here at NJIT on various >computers. >The benchmark is a compute bound matrix solution to find the capicatence of >a square plate via point charge analysis. The speed killer is to solve a >4x4, 9x9, 16x16, up to a 400x400 floating point matrix. An IBM PC (without >8087) was stopped after a week, it didn't complete. An AT&T 3b5 with an >average load of 5 was stopped after 3 days (system crash). An IBM 4361 >with an average load of 3 users took about 9 hours. A Univac 90/80-4 >with about 80 users took 10 minutes. Need I say more about the power >of mainframes? And the Univac is 1970's technology to boot. About 1978 or so; the 90 series was the final evolution of the RCA Spectre system which Univac purchased in the mid 70's. The 9080 was a case in point vis-a-vis the rationale for the world wanting something like the PC or MS-DOS. The machine's basic problem was that it was too good... it made the Univac 1100 series machines look bad, at least as far as the functionality of the basic system software went; I've never seen any sort of a mips type comparison between the two and that's really not that relevant. The Univac lifers killed the 9080 by seeing to it no modern compilers or later generation software was ever written for it. The last I heard, and this was as recently as 1984, was that no 77 level Fortran or 81 level Cobol existed for the 90's and users were being forced to "upgrade" to 1160s and 1180s. Imagine "upgrading" from a mainframe with paging to one WITHOUT it? Univac must have expected the users to figure that since they were getting screwed this time, they would probably be okay in the future. I mean, the LAW OF AVERAGES probably says they can't get screwed the same way two times in a row, right? At any rate, if I want a modern Fortran compiler for my little Chinese AT, I can simply send $500 to Lahey and get one. And, this summer, there will be 80386 mother-boards out there for under $1000. I'd just about be willing to bet that I could run a 400x400 matrix routine on a system like that in less than the two months it would take the 9080 users to convert the program to F66 or WATFOR and THEN run it. Ted Holden IMS