From owner-freebsd-hardware Wed Sep 11 16:40:20 2002 Delivered-To: freebsd-hardware@freebsd.org Received: from mx1.FreeBSD.org (mx1.FreeBSD.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id 80DB337B400 for ; Wed, 11 Sep 2002 16:40:13 -0700 (PDT) Received: from postal1.lbl.gov (postal1.lbl.gov [128.3.7.82]) by mx1.FreeBSD.org (Postfix) with ESMTP id C3D4643E42 for ; Wed, 11 Sep 2002 16:40:12 -0700 (PDT) (envelope-from j_guojun@lbl.gov) Received: from postal1.lbl.gov (localhost [127.0.0.1]) by postal1.lbl.gov (8.11.2/8.11.2) with ESMTP id g8BNeBN24013 for ; Wed, 11 Sep 2002 16:40:11 -0700 (PDT) Received: from lbl.gov (gracie.lbl.gov [131.243.2.175]) by postal1.lbl.gov (8.11.2/8.11.2) with ESMTP id g8BNe9q23984; Wed, 11 Sep 2002 16:40:09 -0700 (PDT) Message-ID: <3D7FD444.7B89FE2F@lbl.gov> Date: Wed, 11 Sep 2002 16:39:48 -0700 From: "Jin Guojun [DSD]" X-Mailer: Mozilla 4.76 [en] (X11; U; FreeBSD 4.5-RELEASE i386) X-Accept-Language: zh, zh-CN, en MIME-Version: 1.0 To: Charles Sprickman Cc: freebsd-hardware@FreeBSD.ORG Subject: Re: Feedback on Intel server SRMK2 References: Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit Sender: owner-freebsd-hardware@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.org Charles Sprickman wrote: > Hi, > > I'm curious if anyone here has used one of these with FreeBSD: > > http://support.intel.com/support/motherboards/server/srmk2/ > > Seems like a nice server, specs look good, lots of used ones available > that will take dual PIII-1.2's. A little out of date, but for $300 > (empty), not too bad. > Have not used this. For $300, you can get dual AMD or single P4 M/B + 1 CPU. > > General opinions on Intel servers/mobos also welcomed. Generally, Intel PCI chipsets always have some problem in either memory controller, L2/L3 cache, or bridge/bridge design. Below compares different server chipsets. Intel 860 had really bad handshaking on Memory Controller Hub (MCH 82860 chipset). It limits the maximum I/O to 90 MB/s. No fixing was posted in Aug. By tuning 82806AA PCI 64 Hub (P64H) chipset, you can increase it to 117.5 MB/s, which I did for SuperMicro (P4DC6+). I suggested them to revise the BIOS. > adjusting the following P64H register bits: > CNF ­ P64H Configuration Register (D31:F0), Address Offset 50-51h, Bit[2], > Delay Transaction (DT) Depth. Change bit 2 from "0" to "1". > Soft_DT_Timer ­ Soft Delayed Transaction Timer Register (D31:F0), > Address Offset 80h, bits[1:0], 1 bit Soft Delayed Transaction Timer. Change bits [1:0] > from default to a lesser value. Intel E7500 chipset does not have such bug, but the memory bandwidth (545) is much less then the ServerWork chipset (672 MB/s) when compared with same XEON CPU. Two IDE buses are actually one, and the total IDE bandwidth is less 100 MB/s. The TCP Rx rate is about 1.6 Gbit/s and Tx rate is about 1.25 Gbit/s (two SK NICs), which is desired. AMD MPX chipset is good on IDE bus design. I have gotten 190 MB/s on two IDE buses. But, its memory bandwidth is lower around 300 MB/s. So, when doing high-speed network I/O, the maximum TCP rate is about 1.0 Gbit/s for both Rx/Tx (one or two SK NICs). All tests are using the same two ECC/REG DDR PC2100 memory. AMD MP +2000 (1.6667 GHz) CPU seems having the similar performance as the XEON 2.0 GHz CPU in general test. XEON CPU has slightly better floating point performance. -Jin To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hardware" in the body of the message