Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!crdgw1!uunet!systech!bpetry
From: bpetry@systech.bjorn.COM (Brian Petry)
Newsgroups: comp.arch
Subject: Re: 82596CA
Keywords: real-life use
Message-ID: <935@systech.bjorn.COM>
Date: 4 Jun 91 19:06:53 GMT
References: <18057@fincastle.fibercom.COM>
Organization: Systech Corporation, San Diego
Lines: 26

In article <18057@fincastle.fibercom.COM> jjf@fibercom.COM (Jeff Fitzgerald) writes:
>Hi,
>	I'm looking for comments on the Intel 82596CA.  This is there
>latest lan controller.  I used the 82586 many years ago and ran into
>all of the (now well-known) problems.  Needless to say I'm a little
>leary of using this part without some good references.  We would be
>using the part in 32-bit linear mode (flat address space) with the
>'simple' memory structure in a high performance application with the
>controller in promiscuous mode.  I'd be interested in any problems
>or quirks people have run into (in this or other modes).
>

We are using the 82596CA in 32-bit linear mode with the 'flexible' memory
structure and run into one of Intel's published errata.  The symptom is that
sometimes, during a collision, the transmitter locks up the whole chip,
both receive and transmit.  The work-around involves a watch-dog timer.
But Intel claims that the problem will not be encountered if you choose
the 'simple' memory structure.

Other than that problem, though, we have had great success with the performance
of the device.

Regards,

Brian Petry
uunet!systech!bpetry