CPU terminology / Performance Question

  • Thread starter Thread starter Bob
  • Start date Start date
Bob said:
On Tue, 27 Jan 2004 14:15:51 -0500, (e-mail address removed) (Paul) wrote:

<snip excellent post>

Paul:

Thanks for a very informative post. One question on the performance:
Are there any web pages or studies that compare celeron/128k to
non-celeron performance(256+K) performance directly for the same
processor speed? I know that this is an "application" question,
but I'd still be interested in seeing some side-by-sides.


I actually started here:
http://www.intel.com/design/celeron/qit/update.pdf

Confusing enough already. I did find this page, which is great
for getting the tech specs for any processor:
http://processorfinder.intel.com/scripts/list.asp

Thanks,
Click to expand...

This page was recently tailor-made just for you :-)
Our friend "P2B" visits here regularly, and he restores and upgrades
440BX based motherboards.

http://www.tipperlinne.com/benchmark.htm

When a task is processor bound, all the information needed is
contained in the cache. The processor can then crunch at the
core clock rate, without having to worry about memory subsystem
performance. When this special set of circumstances apply, then
a simple comparison of core clock rate, is all that is required
to determine the performance difference.

At the other end of the spectrum, is the "cache buster" application.
Such an application has poor "locality of reference", meaning many
instructions and all the data must be fetched from memory. This is
where the modern processors excel, as the memory bandwidth is much
higher. An example of a "cache buster" would be a logic simulation
for chip design.

Fortunately, the majority of desktop applications aren't "cache
busters", so more cache helps a bit. Like I said, the extra cache
might be worth a couple hundred megahertz of advantage.

Another kind of benchmark would be SPECINT/SPECFP. In the
"old days", you'd check a table like this, to find the
info you wanted. This is now a historical reference...

ftp://ftp.cdf.toronto.edu/pub/spectable

You might try looking at tomshardware.com for some articles that
benchmarked and compared processors. There is at least one article
over there that compared a large number of processor types and speeds.

HTH,
Paul
 
The benchmark program ends up being cached, even with only 128k of
cache, so you don't see as much variation as you probably will with most
real systems running multiple applications under Windows. The
difference is usually greater than 10%, but it's highly application
dependent.
 
It should be noted that with Celeron CPU's, that the On-Chip cache ran
at the CPU rated speed, where with the P-II's that the on-die cache ran
at half the CPU speed.

This was only for the P-II and P-II series/type of Celeron chips, the
second generation Celerons, the one from the P-III/P4 dies I don't have
information on.

So, what this meant was that in several real world cases, that a 500 Mhz
Celeron could easily outperform a 500 Mhz P-II CPU. While the P-II had
twice the cache, it only ran half as fast as the cache on the Celeron of
the same speed.
 
Although your point is accurate, it was not a Celeron vs. Pentium issue,
rather it was because the CPUs with on-chip cache (not withstanding if
they were Celerons or PII's) were from a later fabrication geometry family.

And someone corrected my original post, the Pentium II/III fabrication
geometry family sequence was:

(earliest) Klamath, Deschutes, Katmai, Coppermine, and Tualatin (latest)

I believe that all Katmai and later had the full speed on-chip cache
(both Pentium and Celeron).
 
Barry said:
Although your point is accurate, it was not a Celeron vs. Pentium issue,
rather it was because the CPUs with on-chip cache (not withstanding if
they were Celerons or PII's) were from a later fabrication geometry family.

And someone corrected my original post, the Pentium II/III fabrication
geometry family sequence was:

(earliest) Klamath, Deschutes, Katmai, Coppermine, and Tualatin (latest)

I believe that all Katmai and later had the full speed on-chip cache
(both Pentium and Celeron).
Click to expand...

Getting there...

Katmai P3 had 512KB of half speed L2, 0.25 micron Celerons had none
(except possibly SL2WM, which processorfinder says had 128KB of full
speed, but that could be an error). Deschutes and Katmai were both 0.25
micron, IIRC all 0.25 micron Celerons were Deschutes.

Coppermines and later definitely all have full speed cache.
 
Back
Top