johnw@astroatc.UUCP (John F. Wardale) (11/11/87)
In article <676@zycad.UUCP> kjb@zycad.UUCP (Kevin Buchs) writes: > >I have seen it many times in the Media. The 10-P is rated >at 375 Mflops, but does Linpack at 25. Why the big difference? >Do other supercomputers behave similarly? What does a Cray 2 >do on Linpack? First, any computer, but most especially on Vector and multi-CPU machines, one must *NEVER* confuse "rated" or GARENTEED-NOT-TO-EXCEED figures with real, or obtainable figures. [Single-headed Scalar Processors tend to get much closer to there claims.] As for the Cray-2, it's young compilers, and slow memory seem to make it be about 1/2 to 7/8 of a an X-MP. [The 2 is a big win if you need the BIG memory, or can't support an X (X's weigh many-tons, and the 2 is *MUCH* lighter!!)] The X-MP does 44 MFLOPS on the 100x100 linpack, (coded BLAS, one processor) and 134 on the 300x300 (480 for 4 heads) and 713 for 4-heads on the 1kx1k. ------ The Cray-2 does 147+ MFLOPS on LLNL loop 7, but has an h-mean of about 6 and a low of 1.7 NOTES: THe 100x100 linpak does *LOTS* of pivoting, and is generally POOR at showing PEAK capacity of any machine. [it's NOT as bad as the h-mean for the (now 24) LLNL-loops (or kernels)] * Linpak numbers for Dongarra memo#23, Jan-86 -- John Wardale ... {seismo | harvard | ihnp4} ! {uwvax | cs.wisc.edu} ! astroatc!johnw To err is human, to really foul up world news requires the net!