fay@wpi.wpi.edu (Peter R Fay) (12/21/90)
Jack Dongarra's "Linear Equations" Benchmarks put an IBM 6000 model 550 (41 MHz) doing Linpack at 27 Mflops (not hand-optimized), which is equivalent to a Convex, Cray-1S, etc. Of course, hand-optimized it is "only" 62 MFlops and can't compete with the vector machines. Does anyone have anything bad to say about it's Flop performance? Is this benchmark representative of its performance on other benchmarks? Granted the 6000 is a different animal than the vector machines, but does anyone buy it anyway for heavy Flop applications because of it's cost or because its less essential to hand-optimize your code in order to get decent performance out of it? -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~ Peter Fay College Computing Center ~ ~ (fay@wpi.edu) Worcester Polytechnic Institute ~ ~ (508)831-5725 100 Institute Road ~ ~ Worcester, MA 01609 ~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
jet@uh.edu (J. Eric Townsend) (12/27/90)
In article <12380@hubcap.clemson.edu> fay@wpi.wpi.edu (Peter R Fay) writes: >Does anyone have anything bad to say about it's Flop performance? Is As long as you aren't trying to do *anything* else (like run a window system, serve files, print documents, etc) you'll get good performance. Unfortunately, the 6000 seems to go catatonic while it context switches. I don't know where this problem lies (hopefully it's software and will be fixed in a future upgrade). The RT-based 8CE is rumored to be the model on which a new "parallel 6000" beast will be built. If this is the case, I might have to actually give in and say I like an IBM product. :-) -- J. Eric Townsend Internet: jet@uh.edu Bitnet: jet@UHOU Systems Mangler - UH Dept. of Mathematics - (713) 749-2120 "If you are the system administrator and this is the first time you are logging into your system, use the login name root." -- IBM RS/6000 docs