Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!tut.cis.ohio-state.edu!ucbvax!hplabs!hp-pcd!hpsgpa!plim
From: plim@hpsgpa.HP.COM (Peter Lim)
Newsgroups: comp.unix.i386
Subject: Re: Cyrix FasMath CX-83D87 80387 replacement with 386/Unix
Message-ID: <830002@hpsgpa.HP.COM>
Date: 1 Dec 89 02:46:39 GMT
References: <1989Nov29.215150.53@jdyx.UUCP>
Organization: HP Singapore IC Design Ctr
Lines: 35

Just curious.

4 to 10 times faster than what ? A machine without Intel's 387 or
a machine with Intel's 387 ?

I know that a 25MHz no cache machine with Intel's 387 whips out close
to 1.2 Million Whetstones per second. Rumor have it that a 486 system
running at 33MHz might be able to whip out 6 Million WPS (for your
info, the HP9000 / 370 -- the current top of the line HP workstation
produces ~ 2 Million WPS). So, I don't think any '387 replacement will
be able to do 4 to 10 times that of an Intel '387 for that will means
4.8 to 12 Million WPS ! Which means a 25MHz no cache 386 will easily
out perform even the 486 with integrated '387 running at 33MHz ! Just
imagine what a 33MHz cache 386 will do ? From experience, a cache
386 will run '387 at about 20% faster throughput, so multiply by
1.2 * 33 / 25, we get 7.6 to 19 Million WPS ! Woooollll !

I don't have figures to calculate if this is possible given the 
limitation of the 386 data transfer bandwidth to and from the '387.

But prove me wrong if you can and I'll buy the new CMOS 387 for
mainframe performance   :-)   !


Regards,
Peter Lim.
HP Singapore IC Design Center.

      E-mail address:              plim@hpsgwg.HP.COM
      Snail Mail address:          Peter Lim
                                   Hewlett Packard Singapore,
                                   (ICDS, ICS)
                                   1150, Depot Road,
                                   Singapore   0410.
      Telephone:                   (065)-279-2289