harrow@exodus.dec.com (Jeff Harrow, NCSE BXB1-2/E02 DTN=293-5128) (07/15/87)
I've been running some benchmarks (programs to be posted in a few days) on my Mac II with some interesting results. Floating point math seems to be about 6 times faster on the Mac II than on the Mac Plus, which is essentally the speedup anticipated between the 68000 and the 68020. However, this does NOT seem to indicate that the 68881 co-processor is getting into the act. These benchmarks do their math via SANE, and I had been led to believe that System 4.1 would patch SANE to utilize the 68881 if it was there, thereby providing a significant increase in the math speed, yet these benchmark results don't seem to confirm this. Conversly, the same benchmarks run DRAMATICALLY faster on an HD2000, which DOES directly utilize the 68881, yielding more fuel to the assumption that the Mac II's SANE is NOT doing so. Apple: Does System 4.1's SANE on the Mac II utilize the 68881? Jeff Harrow Work address: ARPAnet: HARROW%EXODUS.DEC@decwrl.ARPA Usenet: decwrl!exodus.dec.com!harrow or {allegra,Shasta,decvax}!decwrl!dec-rhea!dec-exodus!harrow Easynet: EXODUS::HARROW Telephone: (617)264-5128 USPS: Digital Equipment Corp. Mail stop: BXB1-2/E02 85 Swanson Road Boxboro, MA 01719
jww@sdcsvax.UCSD.EDU (Joel West) (07/16/87)
You don't say what you're benchmarking; arithmetic, trig, logs, whatever? SANE can use the 881 for arithmetic, but the 881 has a rounding error in the last digit (according to Apple) for the trancendentals. For accuracy's sake, Apple recommends you use SANE rather than the 881, and clearly the Mac II SANE strives for the accuracy of the 128K ROM version rather than the speed of the 881. Don't forget that a sizable amount of the cost of SANE is trap and package dispatching; for that, the linear speedup of 68000 instructions is the payoff. That's why going straight to the 881 is 20-100x as fast as a Plus. -- Joel West, Palomar Software, Inc. (c/o UCSD) {ucbvax,ihnp4}!sdcsvax!jww or jww@sdcsvax.ucsd.edu