iphwk%MTSUNIX1.BITNET@cunyvm.cuny.edu (Bill Kinnersley) (10/09/88)
This concerns the previous discussion about 68881 coprocessor boards for the Amiga and whether they can really give a significant improvement in the execution of programs which are numerically intensive. If you recall the board I have is a 68010/68881 combination from Netch Computer Prods. It runs at the standard Amiga clock speed, although the 68881 can go faster if you add a crystal. I was looking for a meaningful floating point benchmark. Whetstone, I found out, is available in a C version from the National Bureau of Standards. Send mail to nbslib@cmr.icst.nbs.gov with the line: send whetstonec from stones I obtained it and tried it out. I used Aztec v3.4 with 16-bit integers and did three runs: 1) FFP cc whet ln whet.o -lm -lc result: 83,000 WIPS 2) Amiga IEEE cc +fi whet ln whet.o -lma -lc result: 11,500 WIPS 3) 68881 IEEE cc +fi whet ln whet.o -lfpua -lma -lc result: 42,000 WIPS fpua.lib is what comes with the board. I did not try to use mx.lib or m8.lib, only followed the instructions provided by NCP. These were all done from ram: with the relevant libraries preloaded. Timing was done by hand, so no great accuracy is implied or intended. But you can see that the FPU is roughly 4 times as fast as IEEE and twice as slow as FFP. Clearly the quality of the software interface could make a big difference here. I'd be interested in hearing how other FPU boards compare with this. -- Bill Kinnersley Physics Department BITNET: iphwk@mtsunix1 Montana State University INTERNET: iphwk@terra.oscs.montana.edu Bozeman, MT 59717 CSNET: iphwk%mtsunix1.bitnet@relay.cs.net (406)994-6174 UUCP: ...ucbvax!mtsunix1.bitnet!iphwk "This message was packed as full as practicable by modern electronic equipment. Some settling of contents may have occurred during transmission."