*BSD News Article 10897


Return to BSD News archive

Received: by minnie.vk1xwt.ampr.org with NNTP
	id AA885 ; Mon, 08 Feb 93 22:00:59 EST
Xref: sserve comp.unix.bsd:10950 comp.sys.ibm.pc.hardware:43558
Path: sserve!manuel.anu.edu.au!munnari.oz.au!sgiblab!spool.mu.edu!yale.edu!ira.uka.de!Germany.EU.net!gmdtub!bigfoot!tmh
From: tmh@keks.first.gmd.de (Thomas Hoberg)
Newsgroups: comp.unix.bsd,comp.sys.ibm.pc.hardware
Subject: Re: ATI Mach32 and Xfree86 or XS3?
Message-ID: <TMH.93Feb6014407@keks.first.gmd.de>
Date: 6 Feb 93 00:44:07 GMT
References: <1ka60dINNhp3@matt.ksu.ksu.edu> <1993Jan29.052847.11959@netcom.com>
	<1993Jan29.175523.13926@cbnewsj.cb.att.com>
	<1993Jan29.193936.13423@netcom.com>
	<1993Jan29.225541.20260@Informatik.TU-Muenchen.DE>
Sender: news@bigfoot.first.gmd.de
Followup-To: comp.unix.bsd
Organization: GMD-FIRST, Berlin
Lines: 68
In-reply-to: roell@informatik.tu-muenchen.de's message of 29 Jan 93 22:55:41 GMT

Since I am about to go shopping...
Well I'd like to get a really good motherboard graphics card combo
that will keep me happy for a year or so.
Things I know or assume:

ATI Mach8/Mach32: Seems to me the Mach32 isn't all that different from
the Mach8. The ATI VGA Wonder is now integrated on the same die, there
is a hardware cursor, panning registers and the ability to map the
frame buffer on the memory bus (linear-not banked--on ISA this means
main memory tops out at 12MB--not acceptable for Unix + X).
Apart from that the graphics engine seems more or less unchanged
judging from the benchmarks I have run on an 8514 Ultra and an Ultra
Pro (both ISA).

S3: The seem to offer a subset of 8514. The graphics engine is also
accessed via brain-dead/low bandwidth i/o ports. They have integrated
VGA, a hardware cursor and they can map their frame buffer to the
memory bus (banked only?).

ISA/EISA/Local bus: ISA sucks, but it's cheap. Eisa sucks wider, but
it has a few advantages over ISA. Local bus is best in many
applications, but you get electrical problems fast. So I guess you
have to compromise, that is gat a few of each. Since EISA gets you
ISA, I'm looking for an EISA motherboard with two or three VL slots
(one less, if an AT Bus interfaces is already on the MB).

Graphics accellerators on the local bus: Just because the current crop
of accellerators (MachX, S3) doesn't shine on local bus, that doesn't
mean you don't need it to get good graphics performance. Those chips
seem to have 16-bit Interfaces to the CPU and add waits, when you burn
them on the local bus with 50MHz. However X was designed for a dumb
(=passive) frame buffer with a high-speed 32-Bit memory interface. If
your system doesn't conform, you get problems. Accellerator support
means lots of coding work and the loss of a high-speed memory
interface means loss of performance that CANNOT be offset by a
graphics accellerator unless it has (practically) unlimited off-screen
storage (some TIGA cards come to mind). Backing store becomes
practically useless on an ATI 8514 Ultra with ISA: Transferring
pixmaps between memory and frame buffer is extremly slow and repaints
might turn out to be faster. The Ultra Pro (Mach32) has a definite
advantage here: It can map the whole frame buffer linearly into the
hosts address space and can do pixmap/window transfers a LOT faster
than the standard 8514 (the memory transfers might actually be 32-bit
wide on EISA and local bus implementations).

GX: I don't know the Weitek chip. I've read an article on the GX some
years ago. I consider the architecture the best for X. It's completely
memory mapped, RISC like and has at worst the performance
characteristics of a passive frame buffer (the 8514 is a lot slower
than an ordinary SVGA in SOME operations). At best it's limited only
by memory bandwidth, just like all other accellerators.

But there are architectures and there are implementations. The 8514
architecture (also used by the S3) is even worse than the EGA and only
surpassed in stupidity by XGA. Accessing an I/O port from user level
in a 32-Bit protected mode environment (as X does) triggers some of
the worst (=biggest) microcode operations that x86 CPUs are capable
off. Still I was generally pleased by the performance that the ATI
cards were capable to provide. The GX is a far better architecture,
but the original SUN implementation has been outperformed long ago.
---
Thomas M. Hoberg   | Internet: tmh@first.gmd.de
1000 Berlin 41     |           tmh@cs.tu-berlin.de
Wielandstr. 4      |
Germany            | BITNET:   tmh@tub.bitnet 
+49-30-851-50-21   |