laurits@itk.unit.no (Ottar L. Osen) (02/19/91)
I am planning to buy a 386 DX maschine that runs on 25MHz. I have read somewhere that it is suffisient to have 2/3 of the speed on the 387. Is it possible to use a 16MHz 387 on a 25 MHz 386 maschine? Or do I need 20 or 35 MHz? I could not find any tables in the Intel databook. i do realize that higher speed -> more power dissipation. How low can you go with sufficient cooling? Yours Ottar
lhughes@b11.ingr.com (Lawrence Hughes) (03/01/91)
In article <1991Feb19.145750.27779@ugle.unit.no>, laurits@itk.unit.no (Ottar L. Osen) writes: > I am planning to buy a 386 DX maschine that runs on 25MHz. > I have read somewhere that it is suffisient to have 2/3 of > the speed on the 387. On the 286, you need a 287 that runs at least 2/3 the clock rate of the CPU. On the 386, you need a 387 that runs at least THE SAME clock rate of the CPU. Since the 387 came well after the 386, early (and some current) 386 boards have sockets for a 287, which typically is set up for ONE THIRD the CPU clock (e.g. on a 25 MHz 386, you need an 8.33MHz 287, although I'm reasonably sure any 8MHz part would work - I HAVE run with an AMD 10MHz 287 on my 386-DX/25). You also need to match 386SX to 387SX and 386DX to 387DX. > Is it possible to use a 16MHz 387 on a 25 MHz 386 maschine? > Or do I need 20 or 35 MHz? ^^ would you believe 25? > i do realize that higher speed -> more power dissipation. How > low can you go with sufficient cooling? Sorry - don't have any idea on this one... just what kind of cooling are you thinking of? Giant heat sink strapped to the chip? Liquid freon cooling? I don't think forced air is going to be much help, unless it requires gale force warning flags.... anyone else have any ideas on this one? > Yours Ottar I bought a fine 80c287-10 from AMD a while back that I used in my 386-25 for a while (worked fine, but it is now happy in my 286/10 MHz), but have upgraded it to a 387DX-25 (mama mia... thatsa some spicy coprocessor) - they occassionaly show up on the net in the $200 to $250 range. They go for about $450 to $550 from mail order and discount houses. List is about $850. Ouch. Note that x87s have the HIGHEST MARGINS of any (generally available) chips being sold, so bargins can be found. Side note: as of October '90, a new series of 287 and 387 are being marketed, with higher performance than previous series (e.g. the 287XL (??)). The new 387 chips have no "S" in the ten digit part number UNDERNEATH the chip, while the old ones have numbers that start with an "S". The new parts are supposed to be 20% faster than the old ones at the same clock rate (fewer clocks per operation???). Mine happens to be a new one. Anyone like to race me to the next traffic light? :-) Seriously, I would be interested in comparative performance between old 387 and new 387 at the same clock rate. Another side note: I have been developing a C callable math library that uses the full 387 instruction set - the libraries with Microsoft (and Borland) C currently wimp out at 287 level. Also, I'm doing some high performace matrix routines directly in 387 assembly. Fun stuff. For one thing, at least some of the 387 opcodes in masm 5.0 assemble the wrong binary... :-) Anyone know if masm 5.1 has fixed this? Try assembling FSIN and see if you really get some 3 byte code (instead of the right 2 byte code) that really gives you COSINE! That was a real trip to track down... ("That's an interesting approach, but IT'S NOT US..." - Nick Danger: Third Eye, Firesign Theater). Borland TASM seems to handle 387 codes just fine. QUERY: anyone know what AMD's plans are for 387 clones? When I ordered my 287, they hinted that they MIGHT have a similar deal "later" - when asked today, they said "we can't currently say anything about that". They suggested looking in the March '91 Byte. For anyone that DOESN'T know - you can order a clone 80c287 10Mhz (no other variants at this time) DIRECT from AMD at 800-888-5590 for $99 + shipping (brand new - prime parts). They work great. ------------------------------------------------------------------------------ Lawrence E. Hughes | ASK DR. MATH - He knows MORE than you do! Remember, Intergraph Corp. CR1100 | He's not a REAL Doctor: "I have a Bachelor's Degree, Huntsville, AL 35984 | in MATH!" * uunet!ingr!b11!lhughes * (205) 730-4167