Date: Mon, 29 Jun 1998 12:42:11 -0600 (MDT) From: Atipa <freebsd@atipa.com> To: "Christopher G. Petrilli" <petrilli@dworkin.amber.org> Cc: Niall Smart <rotel@indigo.ie>, freebsd-smp@FreeBSD.ORG Subject: Re: PPro vs PII Message-ID: <Pine.BSF.3.96.980629123522.7863B-100000@altrox.atipa.com> In-Reply-To: <Pine.BSF.3.96.980629094802.28663A-100000@dworkin.amber.org>
next in thread | previous in thread | raw e-mail | index | archive | help
> > The P2 will smoke it. Better yet, go up to 350 or 400MHz, then you can > > utilize 100MHz system bus. > > This is a red herring... the system bus was a big restriction back when > the cache was running at bus speed---any increment made a huge > difference, but with the current PII architecture, the bus speed has > long since ceased to be a problem. How many devices do you know that > can saturate a PCI bus constitently? Don't use this as a reasoning. The bus can easily get saturated, and this is a _big_ difference, considering the DRAM, L2, PCI, AGP, etc., all are directly related to system bus speed. You yourself later admit that I/O usually binds system speed, and the system bus is a large portion. You can saturate a PCI bus with any one of the following: o (1) gigabit ethernet o (1) Matrox Meteor o (1) Adaptec 7890 Ultra-2 SCSI With any _one_ of those devices, excluding IDE, normal 100MBit ethernet, anf video (which is _very_heavy_) > > Since the P2 has DIB (dual independent bus) for the L2 cache, higher clock > > rates, and much faster DRAM access, you'll definitely notice the > > difference. Pros are at the end of their lifecycle, and will be hard to > > support. > > The PROs will be faster ata given clock speed, and with the cache > architecture of the Pros, probably at 50% above that, given a normal > model of execution (80-90% cache hit rate). Remember, that you can get > PPros with 512K or 1Mb of L2 cache that is running 1:1 with the chip, > rather than 2:1. But we are not talking philosophy here; who cares about clock-for-clock? This is the real world, where net performance is more important that theoretical (clock-for-clock) benchmarks. > The big problem in most systems in my view (having been responsible for > system optimization of mainframes and large UNIX boxes) is never the > CPU---unless you're doing scientific applications, then it's usually > memory---it's almost always I/O bandwidth. A faster disk, more memory > for caching, more SCSI busses, etc, will make a BIG impact on your > overal system throughput-- which a 100MHz bus will improve by 50%... > -which is a much better measure than > "boboMIPS" :-) Just cuz you're CPU can spin faster, doesn't mean it > doesn't do anything useful during those cycles. Correct. I would invest in a nice CCD setup w/ the Adpatec Ultra-2. But this is the -SMP list, so these are people that DO NEED the extra CPU. Probably breaking keys... ;) Kevin To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-smp" 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.3.96.980629123522.7863B-100000>