From owner-freebsd-smp Tue Apr 29 00:04:45 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.5/8.8.5) id AAA00753 for smp-outgoing; Tue, 29 Apr 1997 00:04:45 -0700 (PDT) Received: from jumping-spider.aracnet.com (root@jumping-spider.aracnet.com [204.188.47.14]) by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id AAA00743 for ; Tue, 29 Apr 1997 00:04:40 -0700 (PDT) Received: from cbrowni2-home (ppp-r21.aracnet.com [205.238.13.23]) by jumping-spider.aracnet.com (8.8.5/8.8.5) with SMTP id AAA10734; Tue, 29 Apr 1997 00:05:14 -0700 Message-ID: <33658F3C.86D@earthling.net> Date: Mon, 28 Apr 1997 23:03:40 -0700 From: Chris Browning Reply-To: brownie@earthling.net X-Mailer: Mozilla 3.01 (WinNT; U) MIME-Version: 1.0 To: Ben Black CC: smp@freebsd.org Subject: Re: Quad Pro 150 motherboard? References: Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Sender: owner-smp@freebsd.org X-Loop: FreeBSD.org Precedence: bulk Ben, > i think a good rul eof thumb is about 256K per 64MB RAM (except for > servers or where bus speeds far exceed RAM speeds). Hmm, I don't quite buy this. First off, what does cache size have to do with memory size? There may be some "general" corrilation in a uniprocessor system, but that is less of a concern in MP environments. The important factor is bus contention, since you have 2-4 (in PPP's case) high speed processors all fighting over the same bus. I believe it is more the case that you want your cache size to grow as a function of the # of processers and with a bus utilization factor thrown in there. Also, the bus speed in servers is typically the same as all other systems, so I don't get the comparison to RAM speeds. Commodity servers use the same RAM speed and bus speed as desktop machines. > so, 4 CPUs with 256K > cache should be fine in non-server configs. the 512K P6 chips are just Well, 4 CPUs IS currently a server configuration. As I mentioned before, all the >2way systems I know of are servers. Granted, these can also make kick-butt workstations, that is not what they were designed for. In addition, 256k is not enough cache for a PPP system for a 4 way system no matter what "configuration" it is. You will get a performance hit because of the bus contention. > not cost effective. last week i ordered a P6-150 for $165. tough to > beat that. That is a good deal. Of course, you loose about 10% of your bus bandwidth running the bus at 60MHz instead of 66MHz (which is what the 150 MHz runs at). > > On Sun, 27 Apr 1997, Chris Browning wrote: > > > Michael, > > I concur with Ben here. Typically in the PC world, anything above > > 2 procs is geared towards servers. I can't think of many non-server > > > 2way systems out there. So, if you want >2way, you will have > > to go with one of the server class machines, which are typically > > expensive. In addition, I would not bother with the 150MHz PPP. If > > I remember correctly, the 150 only comes in the 256k cache size. If > > you are going to do 4way, do yourself a favor and get the 512k cache > > PPP. 4way PPP will saturate the processor bus quite quickly, so the > > more cache the better. I believe they make a 166/512k PPP, so for > > cost effectiveness, that is what I would recommend. > > > > Chris > > Not speaking for Intel. > > Ben Black wrote: > > > > > > a 4 or 6 CPU P6 board for *other* than a large server...think about that > > > for a few minutes. > > > > > > (hint: NT4 Workstation can't handle more than 2CPUs and that is the most > > > popular commercial SMP-capable OS) > > > > > > On Mon, 28 Apr 1997, Mr M P Searle wrote: > > > > > > > Is there such a thing as a cheap quad Pentium Pro motherboard? I'm looking > > > > for an SMP Pro 150, but there aren't many 4 or 6 Pro motherboards > > > > around, and those that I could find were for large servers (eg Intel Alder, > > > > Goliath, etc.) > > > > > > > > Thanks, Michael.