From owner-freebsd-hackers  Tue Dec 16 04:01:48 1997
Return-Path: <owner-freebsd-hackers>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.7/8.8.7) id EAA20073
          for hackers-outgoing; Tue, 16 Dec 1997 04:01:48 -0800 (PST)
          (envelope-from owner-freebsd-hackers)
Received: from arg1.demon.co.uk (arg1.demon.co.uk [194.222.34.166])
          by hub.freebsd.org (8.8.7/8.8.7) with ESMTP id EAA20067
          for <hackers@freebsd.org>; Tue, 16 Dec 1997 04:01:40 -0800 (PST)
          (envelope-from arg@arg1.demon.co.uk)
Received: (from arg@localhost)
	by arg1.demon.co.uk (8.8.5/8.8.5) id LAA12880;
	Tue, 16 Dec 1997 11:59:19 GMT
Date: Tue, 16 Dec 1997 11:59:18 +0000 (GMT)
From: Andrew Gordon <arg@arg1.demon.co.uk>
X-Sender: arg@server.arg.sj.co.uk
To: John Kelly <jak@cetlink.net>
cc: FreeBSD Hackers <hackers@freebsd.org>, Mike Smith <mike@smith.net.au>
Subject: Re: 3com 3c509 card 
In-Reply-To: <34964c48.30733173@mail.cetlink.net>
Message-ID: <Pine.BSF.3.91.971216115308.12854C-100000@server.arg.sj.co.uk>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-freebsd-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

On Tue, 16 Dec 1997, John Kelly wrote:

> On Mon, 15 Dec 1997 23:51:31 -0800 (PST), Tom <tom@sdf.com> wrote:
> 
> >  Also, the ed driver supports shared memory cards like the SMC Ultra 16.
> >Fastest ISA ethernet you can get.
> 
> I replaced an NE2000 clone with a SMC Ultra 16, thinking shared memory
> would consume far less CPU than PIO with an NE2000.  But it seems to
> be about the same.

Have you got a _real_ (no longer available) Ultra here?  Testing with FTP 
(on a 486), I get about 600-700kbyte/sec with NE2000s, and 900-1000kbyte/s
with SMC Ultra.  However, the "EtherEZ" which has replaced the Ultra 
appears to be a bit of a lemon - giving rather inconsistent performance
(dropped packets) and even when not dropping packets, worse throughput 
than the Ultra.  This appears to be due to its reduced buffer size (8K 
vs. 16K); also, I notice that it only has one chip for the 8K RAM (vs. 2 
chips on the Ultra) - so, although still notionally a 16-bit card, I 
suspect it might be multiplexing accesses to a single 8-bit wide device 
and hence needing to insert wait states.  I haven't actually proven this 
theory, but the performance is certainluy not good.

(above tests done with several examples of each type, so I'm not 
suffering from faulty cards).