Date: Sat, 12 Sep 1998 12:27:11 -0400 (EDT) From: spork <spork@super-g.com> To: freebsd-hardware@FreeBSD.ORG Subject: Re: "Cacheable memory"?? (fwd) Message-ID: <Pine.BSF.4.00.9809121226110.9027-100000@super-g.inch.com>
next in thread | raw e-mail | index | archive | help
I checked with the author and he OK'd posting this, it's a good answer... Charles ---------- Forwarded message ---------- Date: Sat, 12 Sep 1998 11:13:48 +1000 From: Kevin Lam <kevla@studentmail.dis.unimelb.edu.au> To: spork <spork@super-g.com> Subject: Re: "Cacheable memory"?? At 15:29 9/11/98 -0400, you wrote: >Pentium II: >Less L2 cache (512K max) >Good, stable ASUS mainboards available (we've had excellent results with >Asus so far) >100MHz bus Pentium II - less L2 cache, however, far greater cacheable memory range, up to a full 4GB on the newer 100Mhz (official bus speed) series chips. Overclocking a 333 on a 100Mhz FSB won't get ya this ;) For some strange reason the VIA MVP3 only seems to be able to cache up to 256MB of RAM with a 1MB L2 cache - crippling, if you ask me, hence the full-blown Pentium II always wins hands down over AMD for 98% of my commercial server applications. >AMD K6-2: >No time-tested mainboards available (the Asus model only does 384M and >only offers 512K cache) Very good point here. No time tested boards and generally, manufacturers have concentrated far more effort in the PII board line, many PII boards are built to server/workstation reliability standards, whereas the impression I get from the S7 market is "consumer grade". While I believe S7 still exists to compete against Intel and that is a good thing, and I still deploy S7 boards at home, I wouldn't bet my server's future on them. >Now how much difference does the L2 cache make in a typical web/mail/news >server? What is meant by the term "cacheable memory"? ie: "with 512K >cache you have 64MB cacheable memory" or "with 1M cache, you have 128MB >cacheable memory". I've also heard things like "this motherboard can only >cache 64MB of memory"... You take performance hits once you go over the limit, ie. putting 128MB of RAM in a 64MB-limit box leaves 64MB cached, and 64MB uncached. Whenever the CPU needs data from the part of RAM which is uncached, it results in the "Celeron Effect" ;) >What does it mean? What's the real world impact? Significant enough that I wouldn't deploy S7 or a Celeron in a production server environment. Performance takes a dip. -- K To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hardware" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.00.9809121226110.9027-100000>