P4 Prescott 3.2GHz

In article <pcSqc.25754$gr.2284479@attbi_s52>, "DaveW" <none.DeleteThis@zero.org> wrote:

 > The Prescott line does indeed run that hot. It has been reviewed on many
 > professional review web sites, and the authors are all saying to get a
 > Northwood for now until Intel solves the heat problem. Adding injury to
 > insult, yes the Prescott's run somewhat slower than Northwoods at this stage
 > in their design. That is supposed to be resolved later this year when the
 > faster clock speed, thermally redesigned Prescotts appear. Oh, and the 875
 > chipset is the fastest and most stable to try running a Prescott on.

I've read a couple of articles in the last week, and the issue is,
at 90nm feature size, there is an increased effect called "DC leakage
current". I hadn't realized just how bad this had become.

Normally, all of the current in a CMOS gate, is used to flip states
on the output. The energy used is 0.5*C*V**2 and the power is
proportional to F*C*V**2. At one time, all that mattered was
reducing C, by making the gates smaller. At the gates get smaller,
the voltage they operate with is reduced (not sure if this is
a breakdown voltage issue, or just the need to reduce power caused
by also attempting to increase F at the same time).

As the voltages drop, the transistors can no longer be completely
turned off by the logic signals coming from the previous stage.
One article I just read, referred to "high threshold" and "low
threshold" transistors, implying that the latter ones waste power
via leakage.

It seems that we are headed for maybe 25% of the power wasted
as heat and doing no useful work. The industry may in fact be
giving up on simply increasing frequency, and instead looking
at multiple cores, more parallelism and so on, as further shrinkage
using the current material science will only make leakage worse.

As for the Prescott performance, I wouldn't give up on the Prescott
until there is a decent compiler developed for it. Any time there
are architecture changes, it takes a while for the changes to be
digested and incorporated into popular compilers. The P4 looked
pretty bad after its introduction, and a tweak here and there might
bring the Prescott back.

I've taken the liberty of extracting all the Intel D875PBZ motherboard
results from (beware, this page is a browser crusher, 1.7MB):

<a style='text-decoration: underline;' href="http://www.specbench.org/cpu2000/results/cpu2000.html" target="_blank">http://www.specbench.org/cpu2000/results/cpu2000.html</a>

Columns are frequency, CINT2000_Base, CINT2000_Peak,
processor+compiler_info

3.20 1583 1620 3.2ee Intel C++ 8.0
3.20 1475 1509 3.2ee Intel C++ 7.1
3.20 1583 1620 3.2ee Intel C++ 8.0
2.40 1039 1071 2.4C Intel C++ 8.0
2.60 1104 1138 2.6C Intel C++ 8.0
2.80 1166 1204 2.8C Intel C++ 8.0
3.00 1152 1160 3.0C Intel C++ 7.0
3.00 1164 1200 3.0C Intel C++ 7.1
3.00 1226 1265 3.0C Intel C++ 8.0
3.20 1221 1261 3.2C Intel C++ 7.1
3.20 1287 1330 3.2C Intel C++ 8.0
3.20 1282 1329 3.2C Intel C++ 8.0
3.40 1341 1389 3.4C Intel C++ 8.0
3.40 1342 1393 3.4C Intel C++ 8.0
3.40 1666 1704 3.4ee Intel C++ 8.0
3.40 1666 1705 3.4ee Intel C++ 8.0
2.80 1219 1269 2.8E Intel C++ 8.0
2.80 1219 1268 2.8E Intel C++ 8.0
3.00 1292 1345 3.0E Intel C++ 8.0
3.00 1292 1345 3.0E Intel C++ 8.0
3.20 1363 1420 3.2E Intel C++ 8.0
3.20 1363 1421 3.2E Intel C++ 8.0
3.40 1432 1491 3.4E Intel C++ 8.0

Rearranging by speed versus processor type, and extracting the
two entries for each one gives the following (only C++ 8.0
compiler entries used)

Freq ___P4-C___ ___P4-E___ __P4-EE___
2.40 1039 1071
2.60 1104 1138
2.80 1166 1204 1219 1269
3.00 1226 1265 1292 1345
3.20 1287 1330 1363 1421 1583 1620
3.40 1342 1393 1432 1491 1666 1705

In each case, you can see the -E is slightly ahead of the -C
and the Extreme Edition beats them both.

It is possible that all SETI needs is a recompile for Prescott
to beat Northwood. Using the raw data above, you can see
what an influence the compiler has on the results.

HTH,
Paul<!-- ~MESSAGE_AFTER~ -->
 >> Stay informed about: P4 Prescott 3.2GHz 

Leigh-Anne Mills wrote:
 > Hiya
 >
 > I have been given a P4 Prescott 3.2GHz to eval by work.
 >
 > No OVERCLOCKING is happening.
 >
 > I have put it in my 2nd PC which as a ASUS P4P800 mobo with latest
 > bios, that had a P4 3GHz Northwood.
 >
 > Couple of things i have noticed about the Prescott, the temp is
 > bloody high. When idle its 61C / 141F, when 50percent loaded its 69C/
 > 156F, when 100percent loaded its up to 71C / 159F. All recored with
 > ASUS Probe.
 >
 > Also i find it slower than a P4 3GHz Northwood, i run SETI on it, and
 > when it says a unit will take 4 hours 55 mins, but the Northwood
 > would do this in half the time.
The Prescott is slightly slower for SOME things than a Northwood at
the same clock speed, but it's usually never more than 5% (and for some
things it's a couple of % faster, as well). "half the time" does not
sound right, if you're comparing the same work units.
In fact, this benchmark <a style='text-decoration: underline;' href="http://forums.hexus.net/showthread.php?t=14693" target="_blank">http://forums.hexus.net/showthread.php?t=14693</a>
seems to indicate it's slightly faster clock for clock for SETI (it's an
app which obviously likes more cache, if you look at the P4EE scores).
The only thing I can think of why it would be only half as fast would be
thermal throttling - but that shouldn't happen at 71C. Or maybe
misdetection of the cpu by the application, though if that would be the
case there surely would be an update available since some time.

 > What is the best chipset to run a prescott on, intel 865 or 875 ??
875 is slightly faster (unless your 865 has the inofficial
PAT-called-differently setting in the bios), doesn't matter if it's a
northwood or prescott. It doesn't make that much of a difference however.

 > Is ther prescott supposed to run this high Temp ??
Unfortunately, yes. The thermal/power characteristics of the prescotts
are less than stellar. Maybe later stepping will improve this a bit, but
for now you're better off with Northwoods - basically the same
performance, but run cooler (and thus easier to cool silently).

 > Is there any settings i should make to bios to speed it up (non -
 > overclocking) ??
Make sure the board is running in dual-channel mode (if you have 2 ram
sticks). Other than that, you could play with ram settings and the like
(though that's also a form of overclocking), but it's probably not worth it.

Roland<!-- ~MESSAGE_AFTER~ -->
 >> Stay informed about: P4 Prescott 3.2GHz