From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Tue, 6 Aug 2002 01:31:37 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Tue, 6 Aug 2002 01:31:37 -0400 Received: from leela.mcs.anl.gov ([140.221.8.226]:10909 "EHLO leela.mcs.anl.gov") by vger.kernel.org with ESMTP id ; Tue, 6 Aug 2002 01:31:36 -0400 Date: Tue, 6 Aug 2002 00:35:08 -0500 From: Robert Latham To: linux-kernel@vger.kernel.org Subject: tigon3: 2466 and bad performance at 32bits/33mhz ? Message-ID: <20020806053508.GJ25554@mcs.anl.gov> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org I've got a 3com 3C996B-T (copper) and a tyan 2466 motherboard. The kernel is 2.4.19-rc1-ac2, and i'm using the tg3 v0.99 (Jun 11, 2002). When i run the 3C996B-T in the 64-bit/66-MHz slot, i get great performance. When i run the card in the 32-bit/66-MHz slot, i get pretty bad performance. Here's a netpipe graph to demonstrate: http://www-unix.mcs.anl.gov/~robl/storage_box/tg3-32_33.2.png The fast curve is with the 3c996B-T in the 64/66 slot. It peaks out around 850 Mbps. The slow curve is the same card in the 32/33 slot. It peaks around 180 Mbps. Is a maximum throughput of ~180 MBits/sec all that can be hoped for? I don't have the hardware to verify, but http://www.cs.uni.edu/~gray/gig-over-copper/#3Com%203c996BT:|outline offers the datapoint of 436 MBits/sec in an oldish dell optiplex ( old being whenever p3 500 mhz machines were still made) So, am i looking at a hardware limitation? driver quirk? I'm open to any suggestions. I'll keep an eye on the web archives, but please CC me on any replies. thanks ==rob -- Rob Latham Mathematics and Computer Science Division A215 0178 EA2D B059 8CDF Argonne National Labs, IL USA B29D F333 664A 4280 315B