Intel Pentium 4 641 (Cedar Mill) - 65-nm process technology advancing
Benchmarking
Despite the theoretically equal speed, we anyway decided to
run a series of tests to clarify all the doubts on this matter. As
benchmarking suites, we used both synthetic benchmarks and real
applications: games and other. Their list looks like this:
- 3DMark’03 v.3.6.0
- 3DMark’06 v.1.0.2
- PCMark’05 (v.1.1.0) CPU test
- SuperPI (8M)
- SiSoftware Sandra 2007
- WinRAR 3.60 (beta3)
- FEAR
- Quake4
To test the processor, we assembled a test bench which would
be able demonstrating all the overclocking capability of the processors
because migration to a finer process technology promises a lot of
surprises for the overclocking fanciers.
|
Test-bench
configuration
|
|
Processor
|
Intel Pentium 4 641 (Prescott, 3.2 GHz, 2 MB cache,
revision B1)
Intel Pentium 4 640 (Cedar Mill, 3.2 GHz, 2 MB cache, revision N0)
|
|
Cooler
|
GlacialTech Igloo 5700MC
|
|
Motherboard
|
ASUS P5LD2 (i945P Express)
|
|
RAM
|
2 x 512 DDR2 Samsung PC5300
|
|
Video Card
|
256 MB Sapphire ATI Radeon X1800GTO ViVo
|
|
HDD
|
120 GB Seagate SATA II, 8 MB cache (ST3120813AS)
|
|
Housing
|
ThermalTake Xaser III (window, 5 housing fans 80-mm)
|
|
PSU
|
FSP Optima 600W (OPS600-80GLN)
|
The first thing to do, we'll verify the performance of the
whole system in famous gaming benchmarks by Futuremark.
There was a good reason to choose 3DMark’03 and
’06 - 3DMark’03 is highly responsive to the change
in the system performance, whereas 3DMark’06 is sort of a new
performance standard. As is seen on the graphs, however sensitive these
benchmarks are to the system performance, the results for both
processors are almost equal.
The task of the second stage is more specific - to run the CPU
test in 3DMark:
Again, the results are almost identical.
We again turn to Futuremark benchmarking suites, this time to
the famous PCMark’05.
The difference in results is most likely caused by the errors,
but with a bias towards the Cedar Mill. We are going further and trying
to calculate the "PI" number with SuperPI at 8 mln digits precision -
the fewer seconds it takes for the processor, the better result is:
Again we see a parity. Trying another most popular synthetic
benchmarking suite - SiSoftware Sandra 2007 - which offers very
convenient arithmetic and multimedia CPU tests. The produced results
have proved expectedly familiar:
All is the same - just minor chaotic discrepancies. At the
final stage, we run a few tests in real applications. For instance,
archiving - a typical task for the computer:
Let's not forget that computer games are known to be the most
popular among "demanding" tasks for the home computer.
Whistle! Both rivals are at the pedestal, and we have to admit
that friendship has won. In fact, we knew that from the very start. The
Cedar Mill core was not supposed to bring any changes to the
performance but merely to raise the bar of maximum operating
frequencies and improve the heat characteristics of Intel Pentium 4
processors.
First, let's talk about the heat characteristics. Of course,
we were mostly interested in finding out about the heat emission of
both the processors, but after producing absurd results we had to run
the tests repeatedly and look for the cause. The cause proved so banal
and sad that you lose heart. Formerly, when testing coolers we
mentioned that the non-uniformity of the heat-spreading lid of the
processor may spoil the whole cooling process from scratch, so whatever
cooler you fit on a faulty processor - that will be of little use.
This time, the situation turned out to be similar. To verify
the uniformity of the base, we used the simplest "old-world" method: a
superfine layer of thermo paste is applied on the glass surface so that
it is translucent, and then the processor being tested is placed with
its "belly" up and pressed from the top. In the end, on the
heat-spreading lid of the processor there remain distinct traces in the
points of contact with the smooth glass surface.
Look what has been produced based on the totals of
verification for Pentium 4 640 and 641 under study:
To the left – Intel Pentium
640, to the right – Intel Pentium 641
The lid of Intel Pentium 640 is arched over the center, and in
Intel Pentium 641 it is concave over the center. Remember that just
under the central part of the lid there is the processor core, that is,
this part is the "hottest" point and the most important is the maximum
contact with the cooling system. Hence, Intel Pentium 640 and Intel
Pentium 641 are in equal conditions, so it is hard to compare their
heat characteristics in such conditions.
Search for information in the Internet allowed us to make a
conclusion that the power consumption of the processor on the new
"finer" Cedar Mill core has gone down by approximately 20% versus
Prescott 2M. Therefore, the heat emission should go down accordingly.
Once there is no way to compare heat emission of processors,
there is a proven method to verify the «dividends»
of the new process technology – the clock speed increase
margin should increase essentially, and overclocking may reveal that.
Now it's now time to move from tests on to overclocking.
Overclocking
While describing the test bench, we hinted that its
configuration had been specially selected in view of overclocking
features, so we came across no obstacles for overclocking. The key role
of the processor cooler was played by the known GlacialTech Igloo
5700MC cooler which has repeatedly shown itself as a leader among
processor coolers, and the housing ThermalTake Xaser III provided
superb ventilation of the whole system.
First, we started squeezing the most of the "old guardian"
Intel Pentium 640. We have to admit we received quite a successful
processor specimen: without raising voltage, it easily overclocked to
3900 MHz, and on raising voltage to 1.55 V we were able to achieve a
stable operation at 4200 MHz. We can state that 4200 MHz is rather
standard overclocking for the Prescott core in the conditions of good
cooling.
We got round to overclocking Intel Pentium 641 with enthusiasm
and expected some "miracle", and fortunately our expectations came
true. We first wanted to overclock the processor without raising
voltage starting from the 3900 MHz bar, but we dared for 4200 MHz from
the start. Imagine how surprised we were to see that Intel Pentium 641
at this clock speed was able to boot up the system and ran all the
stability tests without a problem. On further raise of frequency we
came across errors in the tests, so we had to acknowledge 4200 MHz as
the maximum stable clock speed.
The new Cedar Mill core allows to overclock the processor
without raising voltage to frequencies which were formerly possible to
attain only through raising voltage and with good cooling. An
impressive result!
But the potential is not yet exhausted, so we moved on to
raising voltage on the processor to 1.55V. In view of not the most
successful topology of the heat-spreading lid, the mode of cooling the
processor proved quite poor, which is especially seen upon raising
voltage - the temperature was going up sharply.
Nevertheless, we were able to boot up easily and make the
system run stably at the fantastic clock speed 4700 MHz, and on further
minor raise in voltage we even were able to start up the system at 4800
MHz!
Unfortunately, we failed to verify the operational stability
of the system at this clock speed – the sharp rise of
temperature upon starting the test did not make it possible to run the
test to the end, but we can state with confidence that for our
processor specimen the maximum stable clock speed on raising voltage to
1.55 V is within 4700…4800 MHz, which is indeed impressive,
isn't it?
You can imagine the shining eyes of overclockers, but this joy
had to be overshadowed because to attain such high results you have to
have a motherboard which is able raising the frequency of the clock
generator from 200 MHz to 320-330 MHz. It is important not only to
raise voltage and clock speed but to operate stably at these
frequencies because in so doing the FSB speed exceeds 1200...1300 MHz.
Far not every motherboard can afford that. Just a piece of advice to
beginner overclockers: to get over higher FSB speeds, it is a good idea
to slightly raise voltage on the North Bridge and take care of its
proper cooling.
Conclusion
Summing up the results of tests, we can state with confidence
that Pentium 4 641 (Cedar Mill) has indeed proved successful. But its
"hobby-horse" is not in the performance but the superb overclocking
capability.
If you intend to use the processor in the nominal mode, then
the most promising choice for you would be to buy its dual-core brother
which is priced 25$ higher.
If you plan overclocking, the situation is changing. The
overclocking capability of a single-core processor on the Cedar Mill
core is much higher than that for the dual-core Presler. Modern
applications have not yet migrated to computational multithreading,
which means that higher clock speed would be of better use than two
cores. Such situation will remain for quite a long time. In this
regard, it turns out that the single-core Cedar Mill processor may
prove a more advantageous choice rather than its dual-core analog.
Besides, we have already mentioned that the single-core Cedar Mill
offers much smaller power consumption and heat emission. At the same
time, soon we will see processors of renewed revision
«Ñ1» which will add the promised power-saving
technologies and let achieve clock speeds as high as 5 GHz!
We'd better forget about Prescott processors - the only thing
what may make them attractive is their very low price. This is most
likely to happen soon - already now only their old stocks are selling.
This is especially topical in view of the forthcoming release of the
new Conroe core, which should make the prices for single-core
processors go down by 8% to 54%.
As regards the "Cedar Mill versus Conroe" comparison, Intel
has run passions so high that their new processor core is depicted as
"devil out of the box". Let's not run into panic until we are able to
verify the capabilities of Conroe in our test lab. Time will come, and
we'll definitely conduct a comparison of performance of all these
processors in action, but for now we confirm the following final
conclusion. If you are up to overclocking the processor, your current
choice would be a single-core Cedar Mill, if you are against such a
"torture", then you can pay extra 10 $ - 30 $ and better get a
dual-core processor built on the Presler core.
We appreciate the computer megamarket
chain Unicom
for the presented and tested hardware.
 |
Top Stories: |
 |
 |
|
MoBo:
| |
|
|
VGA Card:
|
|
|
|
CPU & Memory:
|
|
