Tualatin on P2B Benchmarks?

[P2B]

I have a fair amount of information on my P2B modification website about
hardware modifications for running dual and single Tualatin processors
on Asus P2B series motherboards, but nothing about the performance of
these systems.

I am considering taking some time over the holidays to collect and post
benchmark information, but I'd appreciate some input on what approach to
take to ensure the results are useful to the potential audience.

It seems to me others might find comparisons to processors they are
currently running on their P2Bs interesting, but unfortunately I don't
have any Coppermine P3s on hand - just a pair of Katmai P3 450s (which
are stable up to 4.5 x 133, 600Mhz).

I do have a pair of multiplier-unlocked P3-S processors, which I think
will prove useful for demonstrating the effect of increasing FSB vs.
increasing CPU clock on a board restricted to running SDRAM.

I'm tempted to use Sisoft Sandra for benchmarking because it's free and
includes a database of reference benchmarks for processors I don't have
available - but are better freeware benchmarks available?

If you are interested in Tualatin on P2B performance data, please post
your suggestions for test scenarios and measurement tools.

TIA

P2B

http://tipperlinne.com/p2bmod

 

[Stephan Grossklass]

I have a fair amount of information on my P2B modification website about
hardware modifications for running dual and single Tualatin processors
on Asus P2B series motherboards, but nothing about the performance of
these systems.

I am considering taking some time over the holidays to collect and post
benchmark information, but I'd appreciate some input on what approach to
take to ensure the results are useful to the potential audience.

It seems to me others might find comparisons to processors they are
currently running on their P2Bs interesting, but unfortunately I don't
have any Coppermine P3s on hand - just a pair of Katmai P3 450s (which
are stable up to 4.5 x 133, 600Mhz).

I could offer two Cel-300A on a P2B-D and a PIII-800EB on a P3B-F for
comparison. NT 5.x.

I do have a pair of multiplier-unlocked P3-S processors, which I think
will prove useful for demonstrating the effect of increasing FSB vs.
increasing CPU clock on a board restricted to running SDRAM.

I'm tempted to use Sisoft Sandra for benchmarking because it's free and
includes a database of reference benchmarks for processors I don't have
available - but are better freeware benchmarks available?

Sandra has one particular drawback: Its benchmarks are 100% synthetic.
The only ones I've found to be of real use/significance are the memory
related benchmarks with buffering turned OFF.

My standard applications benchmark is the Mozilla 1.5 launching time
(with a single empty browser window - when launched from disk for the
first time, this depends on the hard disk, file system, memory and CPU
performance; this can be reduced to just memory and CPU by closing the
lizard again and measuring the time needed for a restart from the disk
cache. Some times measured here:

Machine 1st launch/s restart/s

P2B-D, 2xCel300A, 512 MiB, Cheetah 36ES 8.3 5.25
18 gig on Advance 2941U2W, NT 5.2

(Same with P2L97-DS, 448 MiB used 2 b 13 7 !!!)

P3B-F, PIII-800EB, 384 MiB, 'Cuda ATA n.m. ~2.1-2.2
20 gig on Ultra66, NT 5.0

Dell Lat. CPt C400GT w/ iBX, mCel-400, n.m. 4.5
256 MiB, Travelstar 40GNX 20 gig on PIIX4E,
NT 5.0

With the lizard running, one may find rendering a local copy of a rather
"heavy" web page a decent benchmark - say, this one I visit rather
frequently: <http://users.erols.com/chare/elec.htm>. (This one may
depend on the font and font size chosen, though. I'd recommend measuring
the time for going back to the end of the large page. That's about 14.5
seconds on the Dual Celery, according to my stopwatch. 18px Verdana.)

Stephan (and merry xmas to y'all!

 

[P2B]

I could offer two Cel-300A on a P2B-D and a PIII-800EB on a P3B-F for
comparison. NT 5.x.

Thanks Stephan

Sandra has one particular drawback: Its benchmarks are 100% synthetic.
The only ones I've found to be of real use/significance are the memory
related benchmarks with buffering turned OFF.

My standard applications benchmark is the Mozilla 1.5 launching time
(with a single empty browser window - when launched from disk for the
first time, this depends on the hard disk, file system, memory and CPU
performance; this can be reduced to just memory and CPU by closing the
lizard again and measuring the time needed for a restart from the disk
cache. Some times measured here:

That's along the lines of what I had in mind - measurements that depend
on memory and CPU performance only, with minimal influence from disk or
video, and relevant to real-world performance.

However, I would prefer a benchmark that doesn't require use of a
stopwatch with it's limited granularity and potential for human error -
besides which I don't own one.

Any suggestions for a benchmark that meets the above criteria but has
the computer taking the measurements?

 

[P2B]

I have now run a bunch of Tualatin on P2B benchmarks and posted the results:

http://tipperlinne.com/benchmark

Also linked from the main page:

http://tipperlinne.com/p2bmod

(I had a battle with the javascript, but the graphs display properly at
1024x768 in Mozilla and IE here - YMMV)

I've made no attempt to interpret the results, but IMHO the numbers do a
reasonably good job of representing the subjective performance
improvement experienced when a P2-B series board is upgraded to Tualatin
processor(s).

Suggestions and feedback would be most welcome.

P2B

 

[anonymous]

I have now run a bunch of Tualatin on P2B benchmarks and posted the results:

http://tipperlinne.com/benchmark

Also linked from the main page:

http://tipperlinne.com/p2bmod

(I had a battle with the javascript, but the graphs display properly at
1024x768 in Mozilla and IE here - YMMV)

I've made no attempt to interpret the results, but IMHO the numbers do a
reasonably good job of representing the subjective performance
improvement experienced when a P2-B series board is upgraded to Tualatin
processor(s).

Suggestions and feedback would be most welcome.

P2B

Interesting.

I'm mostly interested in celly 1000/100 at 1330/133 vs celly 1400/100.

P3-s is still too expensive.

I'm finding my p3 733 does all I need right now.

Upgrading to win2k, adding more memory, upgrading hard disks, and
adding a DVD burner seemed like better places to spend the bucks. All
portable except the memory.

Someone, somewhere, always needs the memory though.

here's an interesting comparison:

http://www.duhvoodooman.com/powrleap/PLbench_5.htm

 

[P2B]

Interesting.

I'm mostly interested in celly 1000/100 at 1330/133 vs celly 1400/100.

If someone would like to contribute Tualeron 1400/100 benchmarks, I'd be
happy to post them - my Tualerons are all 10x multiplier.

P3-s is still too expensive.

P3-S is only about 7% faster than Tualeron clock-for-clock so doesn't
justify the price - but there's no choice for dual systems.

I'm finding my p3 733 does all I need right now.

Upgrading to win2k, adding more memory, upgrading hard disks, and
adding a DVD burner seemed like better places to spend the bucks. All
portable except the memory.

Someone, somewhere, always needs the memory though.

here's an interesting comparison:

http://www.duhvoodooman.com/powrleap/PLbench_5.htm

Different boards and benchmarks, but quite similar to my results -
interesting

 

[Vishal Goklani]

Hi,

I think a beneficial benchmark would be a dual 1.4GHZ PIII-S P2B-DS
setup compared to a single 2.4GHZ P4. Unfortunatly, I haven't seen
too many comparisons between any dual P3 setup, and a single P4.

Thanks,

-V-

 

[P2B]

Hi,

I think a beneficial benchmark would be a dual 1.4GHZ PIII-S P2B-DS
setup compared to a single 2.4GHZ P4. Unfortunatly, I haven't seen
too many comparisons between any dual P3 setup, and a single P4.

Thanks,

-V-

I agree - but I don't have a 2.4Ghz P4 system to compare with.

If anyone with such a system would like to run the same set of benchmark
tests (they are all linked from my site) and send me the numbers, I'd be
happy to add them.