[ale] 386/486 sx/dx, harware floating-point

Kenneth W Cochran kwc at world.std.com
Mon Sep 4 08:49:10 EDT 2000


>From: "Frank Zamenski" <fzamenski at voyager.net>
>Subject: Re: [ale] Re: Multi-drop PPPoE (Re: Mind/link DSL and linux?) --   386-486sx/dx
>Date: Sun, 3 Sep 2000 21:33:39 -0400
>
>I hope nobody is minding (much, anyway) a slight hardware
>tangent, I'm rather interested in this stuff myself, even
>the old stuff. Anyway...
>
[... snippage...]
>
>Also, interesting that your 486DX33 tops the P75. Okay, L2
>is disabled, but I've still experienced many 486s easily
>outperforming a P75 (with MS as common OS to both), and IMO
>a 486DX4 subjectively outperforms my old P75 significantly.
>
>-fgz

Not only "subjectively," but quantitatively as well...  :)
"In the early Pentium days" I read that unless you were
running an OS or application that was Pentium-optimized (&
only a *very* few were, mostly CAD/graphics), the only
performance advantage of the Pentuim over the 486 was in
clock-speed.  This became relatively "well-known" & I'm
guessing that this hurt Pentium sales until it went to
133MHz+.  Legend has it that a 486DX4/100 would outperform
a P-120 on "typical" application/system mixes.

Wrt hardware FPUs (example, 387, et al):

Unix-type OSes of yore seemed to "really like" hardware
floating-point.  :)

Back when SCO & Interactive (ATT SVR3.2) were practically the
only things around, I read an article somewhere demonstrating
the (order of magnitude) advantages of hardware FPU even on
(unexpected) things like *awk*.  My guess at the time was
that there were some FPU-interrupt things going on to account
for this, but after learning a few years ago that Perl does
*all* its math as floating-point, I'd guess that this would
be a major factor too.

-kc
--
To unsubscribe: mail majordomo at ale.org with "unsubscribe ale" in message body.





More information about the Ale mailing list