Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uunet!aplcen!samsung!cs.utexas.edu!ico!rcd
From: rcd@ico.isc.com (Dick Dunn)
Newsgroups: comp.arch
Subject: Re: 386 machines are workstations? (Sun/386i)
Summary: dumb frame buffer: memory speed limits 386
Message-ID: <1990May30.041648.9063@ico.isc.com>
Date: 30 May 90 04:16:48 GMT
References: <136288@sun.Eng.Sun.COM>
Organization: Interactive Systems Corporation, Boulder, CO
Lines: 21

david@eng.sun.com  writes:
> If you look at MIPS and memory bandwidth, a 386 class CPU driving a 1 Mb
> dumb frame buffer is pretty marginal.  You have to tune the window system
> code very carefully to get snappy interactive performance, and this was
> never done for SunView on the 386i.

The experience of the folks working on X at Interactive seems to be quite
the opposite:  If you've got a dumb frame buffer, the limiting factor is
the video memory speed.  This was certainly true for both the Sigma
Laserview and the Cornerstone Video Controllers.  I worked on the Corner-
stone server.  In tuning it, I quickly found that after the obvious stuff
(a few days worth), what mattered was anything that would minimize the
number of accesses to video memory.  Remember that a dumb frame buffer is
not just a hunk of memory; it's a hunk of *dual-ported* memory, and the
bandwidth appetite of a high-res fast-refresh display is enormous (by the
standards of such machines).  Most of the memory's bandwidth goes into
feeding the display; the CPU gets the leftovers.  Given that, it's not too
hard to drive it to capacity with a 386.
-- 
Dick Dunn     rcd@ico.isc.com    uucp: {ncar,nbires}!ico!rcd     (303)449-2870
   ...Simpler is better.