On Mon, 5 Oct 1998, Robert G. Brown wrote:
> On Sun, 4 Oct 1998, Robert M. Hyatt wrote:
>
> > A PII/400 is 2x the pro, but I've never seen a PII/300 that is 2x
> > faster... not quite 1.5 in fact, due to the poor L2 cache on the
> > PII.
>
> My 300 MHz PII's execute my Monte Carlo programs (moderately large at
> ~3-6 MB, certainly big enough to use the cache but not flog it) almost
> exactly 1.5 times faster than my 200 MHz PPros. Also, the factor of
> ALMOST 2 I cited was WRT his proposed purchase of 180 MHz PPros (2x180 =
> 360 compared to 300, within 20%). In my experience, the larger PII
> cache (unless otherwise specified, I'm assuming that he was talking
> about getting the common 256K cache 180 MHz PPro) and SDRAM just about
> exactly compensate for the half-speed cache clock (which, at 150 MHz, is
> STILL almost as fast as a 180 MHz cache clock on a PPro at that speed).
>
> We may be comparing apples and oranges here, though. You are (as I
> recall) running a QUAD PPro with interleaved memory, which has the best
> memory performance available of any PPro architecture. More mundane
> dual CPU motherboards, running ordinary EDO, cannot compete with SDRAM.
> If your benchmarks are memory intensive and based on your quad PPro, I
> wouldn't be at all surprised if a 300 MHz PII is only 1.3x or 1.4x
> faster, instead of the full 1.5. For those kinds of applications, your
> PPros can be 1.1x or 1.2x faster than a PPro on a non-interleaved
> system.
I haven't run any real benchmarks comparing my quad vs a single-cpu
machine, but you might be close to right. However, my 4-way interleaving
(using only FPM RAM, not EDO or SDRAM) probably comes close to running
like SDRAM at 66mhz, because the first word has a long delay, just like
SDRAM, but the next three words pop in one bus cycle apart, again just
like SDRAM with a good chipset supporting it...
>
> > no argument there, but I have a quad P6/200 that will hang right in there
> > with the best dual PII/400's... although the new xeon is quite a machine
> > and will support 4x machines which will definitely toast a quad P6...
>
> Sure -- but which one would you BUY TODAY if offered a choice. The quad
> P6 is what it is and is limited in available memory (although the SPEED
> of its memory is admirable:-). A dual 450 (which is really the fair
> basis of comparison) is far cheaper than a quad PPro (unless prices have
> dramatically dropped since last I checked, always a possibility), holds
> more memory (especially more memory per CPU), holds faster memory, and
> is running its cache clock a bit faster than 200 MHz. Or a dual Xeon
> (which is still mighty pricey:-(. Or, as you say, if you've got the
> bucks a quad or octet Xeon is ultra-cool, although one DOES start to
> wonder how they are going to architect the system so as not to be memory
> bound. Interleaved SDRAM? Is that possible?
>
I would personally buy a dual PII/400 if performance was an issue. The
100mhz bus seems to make it a fair amount faster than my quad-p6, based
on lots of folks running "parallel crafty" on such machines. However, for
what I am doing, I prefer my quad, because I am pretty sure that what runs
3.3X faster on my quad will run 3.3x faster on a quad-xeon as well. IE,
for me, doing parallel algorithm programming, the number of processors is
actually more important than the raw performance...
But if Linus was to send me his quad-xeon, damned if I would refuse it
when UPS walked up to my door. :)
> rgb
>
> Robert G. Brown http://www.phy.duke.edu/~rgb/
> Duke University Dept. of Physics, Box 90305
> Durham, N.C. 27708-0305
> Phone: 1-919-660-2567 Fax: 919-660-2525 email:[EMAIL PROTECTED]
>
>
>
>
