AMD or Intel 80387 Math Coprocessor IC

From: Eric J. Korpela <korpela_at_ssl.berkeley.edu>
Date: Wed Feb 26 20:43:01 2003

> > I'm trying to track down a 387 math coprocessor IC for an old
> 386-based
> > Linux box that's going to be doing a bit of numbercrunching for me. Has
> > anyone here got an AMD or Intel (AMD preferred) 387 coprocessor rated at
> > 40MHz (-40 part number suffix) that would work correctly with an AMD
> > Am386DX-40? No, before you ask, the 386DX does *not* have a built in
> > mathco - the 486DX was (IIRC) the first DX-series chip with a built-in
> > coprocessor.
>
> They come up on epay all the time, pick yer flavor. I just got a 40MHz
> Cyrix Fasmath chip for $2 (includes $1 shipping) a week ago. Supposedly
> clock for clock it's the fastest of the '387 FPUs, but IDK. Popped right in
> to to my P70.

It was a tiny bit faster than the AMD and Intel parts depending upon instruction
mix.

> I'm still looking for a set of eight 2MB 72 pin IBM SIMMs with presence
> detect, and a Cyrix 486Drx2, which is a 486 with a '386 pinout, if anyone
> has a spare.

It's not quite a 486. It actually is a double clocked 386 with a small
internal L1 cache. It doesn't share most of the instructions unique to the 486.

I don't currently have a DRX2, but I do have a DRX, which is essentially the
same thing without the clock doubled core. The L1 cache made it about 20%
faster than an equivalently clocked 386. You had to explicitly turn on the
cache, and set regions as uncachable. (Caching the video card usually made
for problems.)

Eric
Received on Wed Feb 26 2003 - 20:43:01 GMT