Path: utzoo!utgpu!water!watmath!clyde!bellcore!rutgers!gatech!sbmsg1!scbhq!uahcs1!madhat!alvitar
From: alvitar@madhat.UUCP (Phil Harbison)
Newsgroups: comp.sys.ibm.pc
Subject: Re: GATHER and say NO to MCA!
Summary: true PC/AT bus masters are rare
Message-ID: <276@madhat.UUCP>
Date: 24 Jul 88 18:53:11 GMT
References: <42900016@uicsrd.csrd.uiuc.edu> <4420008@hplvly.HP.COM>
Organization: DataVision, Huntsville AL
Lines: 25

In article <4420008@hplvly.HP.COM>, boyne@hplvly.HP.COM (Art Boyne) writes:
> farren@gethen.UUCP (Michael J. Farren) writes:
> 
> > If doing true coprocessing on the PC (AT) bus is so trivial, why haven't
> > we seen cards that work, and work well?
> 
> Have you looked at the "Viper" card available from Hewlett-Packard? ...

Coprocessor  cards for the AT certainly exist, but most communicate with
the  systems  through dual-port RAM or a DMA channel.  I have not seen a
true  bus  master  for  the  AT  bus,  that  is, a board that drives the
address,  data, and control lines after the motherboard tri-states those
signals.   This  is  probably because such an interface is difficult for
the AT, and impossible for the XT.

The  original  PC/XT  bus  had  no multi-master capabilities whatsoever.
This  feature  was kludged into the AT bus, but is difficult to use.  To
get  control  of  the bus, you have to request service on a DMA channel,
wait  for  a  DMA  acknowledge,  then  assert  a  signal that causes the
motherboard  to  tri-state  the  bus.   The  protocol is very cumbersome
compared to most modern busses.

----
Live: Phil Harbison
UUCP: madhat!alvitar