From owner-freebsd-hardware Wed Mar 17 1:21:55 1999 Delivered-To: freebsd-hardware@freebsd.org Received: from alcanet.com.au (border.alcanet.com.au [203.62.196.10]) by hub.freebsd.org (Postfix) with ESMTP id 9DF3314C46 for ; Wed, 17 Mar 1999 01:21:51 -0800 (PST) (envelope-from peter.jeremy@auss2.alcatel.com.au) Received: by border.alcanet.com.au id <40365>; Wed, 17 Mar 1999 19:09:00 +1000 Date: Wed, 17 Mar 1999 19:21:20 +1000 From: Peter Jeremy Subject: Re: P-II vs K6-2 To: richard@pegasus.com Cc: freebsd-hardware@FreeBSD.ORG Message-Id: <99Mar17.190900est.40365@border.alcanet.com.au> Sender: owner-freebsd-hardware@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org richard@pegasus.com (Richard Foulk) wrote: [The Celeron has a small, fast cache compared to the P-II] >I don't consider 64MB to be small. What are you using that has a 64M L2 cache? The biggest I've seen is 4M (on DEC Alpha's) - though I believe some of the newest Alpha CPU modules have 8M. > The old rule of thumb used to be >that 64MB is enough for 80% hit rate, on average. Hit rates vary widely with application, so `average' is very rubbery. >Diminishing returns come pretty quickly after that. Note that a small increase in hit-rate can translate to a large increase in performance. If a main-memory fetch takes 10 ticks and a cache fetch 1 tick, an `average' fetch takes 2.8 ticks with 80% hit rate and 2.35 ticks at 85% hit-rate - close to 20% better performance. (Anyone who has real figures for a normal IA-32 memory hierarchy can feel free to supply realistic numbers). >How big is the Celeron cache? 128K AFAIK. Peter To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hardware" in the body of the message