george@wombat.UUCP (George Scolaro) (07/22/88)
And for some of the newer CISC machines (the NS32532), here are the numbers. This was run on a PC add-in card with a 20 Mhz 32532, running 4 wait states (fast rams are expensive right now, I'll post the faster 1 wait state timing at a later date). In article <1662@gofast.camcon.uucp> anc@camcon.uucp (Adrian Cockcroft) writes: >In article <2294@sugar.UUCP>, karl@sugar.UUCP (Karl Lehenbauer) writes: >> >> Array size 20000 40000 >> Machine (bytes) time (secs) NS32532@20 Mhz 4ws 0.47 0.95 > SUN 4/110 (7 MIPS) 0.61 1.23 > SUN 3/260 (4 MIPS) 0.96 1.94 >> Turbo-Amiga 1.14 2.32 >> VAX 8600 1.19 2.64 >> VAX-11/780 3.04 6.38 >> Amiga 5.68 11.50 >> VAX-11/750 6.11 13.13 >> IBM PC AT 8.13 99.71 > For those not familiar with the NS32532, it is a continuation of the Series 32000 processors from National Semiconductor. Most instructions are 2 clocks, giving a maximum performance of 15 native MIPS at 30 Mhz. A four stage pipeline allows address computation in parallel with instruction execution, and the read-ahead/write-behind logic pemits memory-to-memory operations at almost full instruction execution rate. The sieve was compiled with the CTP compilers from National, and this was not a bogus time - no instructions were "optimized out". More numbers as we get them... George Scolaro & Dave Rand george@wombat dlr@daver.uucp (try {pyramid|sun|vsi1|killer}!daver!dlr !daver!wombat!george)