scotte@locus.com (Scott D. Eberline) (05/21/91)
In article <43523@netnews.upenn.edu> bilston@eniac.seas.upenn.edu (Lynne Bilston) writes: >I have a 385/25 machine, brand new, and I want to buy a maths >coprocessor for it. I have seen ads in PC magazines for >Cyrix and IIT coprocessors, quite a bit cheaper than intel ones. >I was wondering are they REALLY 100% compatible with IBM (clone) >machines, and whether you can just plug them into the existing >slot like an intel coprocessor and have them behave exactly as >an intel coprocessor (except faster I believe, according to the >ads)?? I've been told by a few kind people on the net (thanks, kind people!) that Cyrix coprocessors (probably the 83C87) cannot run asynchronously. This is really only important if you wish to run the coprocessor at a different clock speed that your CPU; for example, your 25 MHz 386 can work with a 387 running asynchronously at up to 40 MHz. Cyrix and IIT coprocessors are supposed to conform to the IEEE floating- point specification, as do the Intel chips. They may not produce exactly the same results in the least-significant bits as Intel for some operations, but this does not mean that they are incorrect -- they may be more precise than the Intel results. But all three should produce results correct as far as the IEEE specifies. A few questions I'd like to add: are Cyrix coprocessors, particularly memory-mapped coprocessors such as the EMC87, enough faster than Intel chips that a Cyrix EMC87 running synchronously at 25 MHz would beat an Intel 80387 running asynchronously at 33 MHz? 40 MHz? Is the EMC87, like the 83C87, incapable of running asynchronously? Does anybody have a phone number for Cyrix? -- Scott D. Eberline scotte@locus.com or lcc!scotte