stevew@en.ecn.purdue.edu (Steven L Wootton) (03/18/91)
I've decided to put together the 80287 Socket Rocket described in the April 1991 issue of ComputerCraft Magazine. This will be done purely for its experimental value. To that end, I need to know a couple of things about 80287 floating-point processesors. First, I would like to use the most efficient 287 out there. There are three major contenders for the project (that I know of): Intel 80287 XL ($185) IIT 2c87-20 ($190) AMD 80287 ($100) Have these three chips been benchmarked against each other at the same clock speed to determine which is the fastest? I remember a BYTE article showing that the AMD chip is slower than the Intel chip, but I don't recall hearing anything about the IIT. Second, I would like to use the chip which can be safely driven at the highest possible clock speed. Adding a heat sink is no problem, and I am planning on doing that anyway. Could any of these chips run at 33MHz? 25MHz? If not, how fast? If a less-efficient chip (like the AMD) could run at a higher clock speed than a more-efficient chip (like the 287XL), then I would probably go with the higher clock speed, if it would yield a greater overall floating-point throughput. Perhaps a better question: how do these things fail if the clock speed is too high? If they tend to catch fire, that would tend to discourage high- speed experimentation :-) All suggestions, hints, info, and questions are welcome. You can post replies or send email; either would be fine. Thanks. Steve Wootton stevew@ecn.purdue.edu stevew@pur-ee.uucp stevew%ecn.purdue.edu@purccvm.bitnet
Virgilio.Vincenzo@p13.f602.n332.z2.fidonet.org (Virgilio Vincenzo) (03/21/91)
I have had the experience with a 80287 intel and an 80c287 by IIT. At the same speed of 10 Mhz, the IIT is 28% faster. But i haven't had experience with a 20 Mhz Math Coprocessor. God Bless Ye. Xiao Vinc. -- WolfNet BBS Pisa (Italy) Tel. +39-50-589050 300-14.4K Baud Matrix 2:332/602.0 Virgilio Vincenzo - via FidoNet node 2:332/602 UUCP: ...!gear!wolf!602.13!Virgilio.Vincenzo ARPA: Virgilio.Vincenzo@p13.f602.n332.z2.fidonet.org
rfutscher@pbs.org (03/23/91)
In article <1991Mar17.222637.4437@en.ecn.purdue.edu>, stevew@en.ecn.purdue.edu (Steven L Wootton) writes: > > I've decided to put together the 80287 Socket Rocket described in the April > 1991 issue of ComputerCraft Magazine. This will be done purely for its > experimental value. To that end, I need to know a couple of things about > 80287 floating-point processesors. > > First, I would like to use the most efficient 287 out there. There are > three major contenders for the project (that I know of): > > Intel 80287 XL ($185) > IIT 2c87-20 ($190) > AMD 80287 ($100) > MUCH DELEATED I have a IIT 10 mhz 287. It is running from a 32 mhz clock that is devided by three for 10 mhz. Because of the devide by three the duty cycle is not 50-50. When I run the tests on it, it fails the modes that that were added by IIT. I think they are matrix functions. The boxes that are drawn on the screen lack x values. As if zero is returned as answers. The boxes are just a series of verital lines centered on the screen. No smoke, no flames. Robert Futscher rfutscher@pbs.org