Date: Wed, 7 Apr 1999 14:24:36 +0930 From: Greg Lehey <grog@lemis.com> To: Darren Pilgrim <dpilgrim@uswest.net>, unknown@riverstyx.net Cc: Mark Ovens <marko@uk.radan.com>, Leif Neland <leif@neland.dk>, FreeBSD Questions <questions@FreeBSD.ORG> Subject: Re: K6-2/333, was: Re: Debug kernel by default (was: System sizewith-g) Message-ID: <19990407142436.D2142@lemis.com> In-Reply-To: <3709EE06.77F97B9E@uswest.net>; from Darren Pilgrim on Tue, Apr 06, 1999 at 04:20:38AM -0700 References: <Pine.LNX.4.04.9904051744500.31071-100000@hades.riverstyx.net> <3709EE06.77F97B9E@uswest.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tuesday, 6 April 1999 at 4:20:38 -0700, Darren Pilgrim wrote: > unknown@riverstyx.net wrote: >> On Mon, 5 Apr 1999, Darren Pilgrim wrote: >>> unknown@riverstyx.net wrote: >>>> I may be out to lunch on this one, but I'm pretty sure that the multiplier >>>> is for the internal clock of the chip. So, if, after applying the >>>> multiplier to one chip you get 300MHz, and after applying a different >>>> multiplier to a different chip with a different bus speed you also get >>>> 300MHz, you get two chips that perform the exact same number of >>>> operations/sec. The difference is the bus speed, which affects I/O >>>> performance, etc. A 100 MHz bus with a x3 multiplier will outperform a 66 >>>> MHz bus with a x4.5 multiplier because the CPU will have to wait more >>>> often when it wants to fetch non-cached data from RAM. >>> >>> While this is mathematically and theoretically sound thinking, tests >>> have shown that there is little CPU/memory performance gain with a >>> 100MHz bus. Just take a look at www.tomshardware.com. As for my own >>> systems, I run K6-2 333s at 5x66 just because it sets the PCI and AGP >>> clocks at their spec'd rate of 33 and 66MHz, respectively, while >>> providing the CPU's spec'd 333MHz. >> >> I've seen good speed gains by moving to a 100MHz bus, although this was >> for servers that were doing a lot of database work and heavy network >> traffic. Perhaps it wouldn't matter much for servers doing more >> calculation-intensive work? > > Aye, in a server setup a faster bus does make a difference, but my > reference (Tom's HW) is for workstations. Did I miss the first part > of the thread, was this discussion about servers? If so, my apologies > for my misunderstanding. No, we're not talking about artificial concepts like servers and workstations, we're talking about the low-level behaviour of the processor and memory. > Disk and memory work in a server can max a slower FSB, but if the > server is being used for CPU-intensive work, then what's the point > of spending extra for a server? I'm not sure I understand this statement. To restate it: the whole matter is relevant only to processor and memory. If your bottleneck is disk I/O, you're not going to see much difference. If you're running lots of long-running instructions such as double floating point divides, the bottleneck is the processor, not the cache interface. If you're running CPU bound with a typical instruction mix (single-cycle execution) and you're accessing a large amount of memory, you will see a significant difference. Depending on what a "server" or "workstation" is doing, any may fit into any of these categories. Greg -- When replying to this message, please copy the original recipients. For more information, see http://www.lemis.com/questions.html See complete headers for address, home page and phone numbers finger grog@lemis.com for PGP public key To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19990407142436.D2142>