Recently, after reading about the higher efficiency of Intel’s 0.13-micron Northwood core P4s, I found myself asking: Can a P4 be made to run silently at 2-GHz using non-exotic, inexpensive techniques? Attempting to answer this question was a great excuse to succumb to the siren call of speed glorious speed – once again. A long, detailed article part DIY, part review, part op-ed. The short answer, BTW, is YES!
April 13, 2002 – by Mike Chin
It is a computing axiom that you can’t have fast and quiet; certainly
not without resorting to elaborate and exotic cooling systems these days. This
is due mostly to the hot multi-gigahertz CPUs from both AMD and Intel, which
have likely caused a average increase in PC noise output of at least 2 dB/year
since 1999, as discussed in this article about noise
emission trends in electronic equipment. Recently, after reading about the
higher efficiency of Intel’s 0.13-micron Northwood core P4s, I decided to put
the axiom to the test: Can a P4 be made to run silently at 2-Ghz using non-exotic, inexpensive techniques?
Attempting to answer this question was a great excuse to succumb the siren call
of speed glorious speed – once again.
In every PC project, one has a strategy, whether improvised or carefully planned
out. My approach this time was mixed. I needed the P4, a motherboard for it,
and DDR memory. But I had enough other components lying around that could be
used. I also wanted to prove at least to myself that making a quiet computer
does not necessarily mean spending a lot of money on specialty products.
I hoped to use an existing case, modified for excellent airflow so that case
fans would not be necessary, with a decent PSU powerful enough for the current
demands of the P4 but not too expensive. Hopefully, the retail P4 heatsink and
fan would be quiet enough to be used. If necessary, I would replace the PSU
and heatsink fans with quieter ones or run them at low voltage.
The P4 and other New Goodies
A retail Intel P4-1.6A was purchased, along with a MSI 645 Ultra 333 (SiS chipset) motherboard and a 256 MB stick of generic PC2100 DDRAM. The P4-1.6A is a 0.13 micron core CPU with 512K of cache introduced just a few months ago. It has been touted as the new Celeron 300A for its extreme overclockability.
First, a picture of the siren herself. The part number is SL668. It was packed
on 01/31/02 and made in Costa Rica. The most surprising aspect of the CPU is
its size. The photo doesn’t show the scale, but it is TINY, measuring just a
bit over 1″ square. Intel calls the metal piece that covers over 90% of the
top surface a heat spreader. The large contact area helps the heat from
core to be dissipated out to the heatsink more efficiently. It also protects
the die from damage during heatsink installation – a problem too often seen
with the AMD socket-A CPUs. The other surprise is the modest price: In late
February 2002, at a local Vancouver shop, it cost just Cdn $235 plus taxes,
which worked out to around US$165 in total.
Initially, I was led to believe that the power dissipation of this P4 at its
rated 1.6 GHz clock speed and 1.5V core voltage is 38.7 watts. That’s the number
touted all over the web: more than 20 watts lower than the equivalent clock
P4 of the previous 0.18 micron Willamette variety, and 10W lower than the power
rating of the AMD T-bird 1GHz in one of my systems.
Later, I learned I was misinformed.
There are 2 versions of the P4-1.6A. According to Intel’s
documentation about the P4, my 1.6A SL668 is a normal 46.8W 1.6A.
At time of writing, all retail box P4-1.6A are the 46.8W variety.
The low wattage 38.0W P4-1.6A is part # SL62S, designated as an OEM part. However,
OEM versions of the Northwood core P4 cannot be found anywhere locally in Vancouver
or on the Internet. Local retailers said the price difference between the two
versions is so small they felt there was no point ordering the OEM, especially
considering minimum ordering requirements.
This meant I was dealing with 9 watts more than expected. Pretty much the same
wattage as my 1 GHz T-bird. Grrrr! Still, I had managed to silence the T-bird,
NOTE: Added June 14, 2002
According to one source, the actual power dissipation of the two P4 versions discussed above is the same; they are the same processor. The lower power version simply can’t handle as much peak power before burning up. Average power dissipation and thermal efficiency is the same.
The Retail P4-1.6A Package
The retail Intel heatsink-fan (HSF) that came with the CPU is more impressive than expected.
The heatsink is fairly large at 2.75″D x 3.25″W x 1.375″H, with 25 fins
on a contoured base that is thicker in the center. The all-aluminum heatsink
weighs ~300 grams, and the fan & mounting clip add ~75 grams. A thermal pad
was attached to the base. I removed it with the help of an old credit card,
then baking soda, dish soap and a sponge with water. With fan detached, of course.
The bottom surface is not mirror-polished like many expensive heatsinks but
smooth enough that I didn’t feel there was any need for lapping.
The fan design is much like that on previous Intel processors where
the housing is a single-piece mold that clips tightly to the HS. Compared to
photos of stock HSF on other P4s I have seen (see the 2nd photo on right), this
fan looks smaller. I believe the fan & the heatsink have been changed for the
0.13 micron Northwood.
The standard core voltage of the NW P4 is 1.5V, compared to 1.75V of the older 0.18-micron core Willamette P4s; power dissipation has dropped by approximately 25% for processors of the same speed. For example, the nominal power spec for the ‘Willamette’ core 1.6G P4 is 61W while that for the P4 1.6A is 46.8W.
So perhaps because of the reduced cooling requirement Intel reduced the fan
size, and perhaps the heatsink? If so, it is unfortunate for HSF silencing,
as smaller diameter fans always whine more while delivering less airflow, and
smaller heatsinks require more airflow for the same level of cooling.
Examining the fan closely, it appears that something of the augmented
fan technology of the Millennium Glaciator and the PBL Panaflo fans may
be in use here. There is little or no frame to speak of, as both in / out flow
sides are almost completely open.
Considering its small 60mm blade size and 3000-rpm speed, it’s not that noisy. It’s rated at 0.16A, made by Sanyo-Denki, and has a part number (109X9412T5H036) that looks like one of Sanyo-Denki’s. I could not find this part on their website nor Intel’s, so can only guesstimate that the CFM is in the mid-20s, and its dBA spec around 30. It is whinier & noisier than my reference quiet fan, the 80mm Panaflo FBA08A12L, rated at just 21 dBA.
The HSF mounting clip is a clever affair that uses two cam action levers to
clamp the heatsink tightly to the motherboard, sandwiching the CPU in between.
The levers are in a plastic frame that comprises half of the mounting clip.
The other half is a plastic frame that is already on the motherboard, around
the CPU socket. It is a bit fiddly with the MSI motherboard because of capacitors
mounted close to one side of the plastic frame. But all in all, HSF mounting
is relatively simple and stress-free, unlike many of the clipped HSF for socket
CPUs, especially with the added security of the protective metal heat-spreader
cover over the core.
An animated GIF shows how this works — not a great sequence, but gives you the general idea. (If the pic below is not animated, hit the browser refresh or reload button — the GIF is set to repeat 10 times, then stop so it doesn’t drive you batty.)
Bowing to Pressure
Still this setup is not without problems. So much pressure is applied that
the motherboard actually gets bowed directly beneath the CPU socket. That’s
right: the motherboard gets bent. It worried me when I noticed it, so I did
some web research to see if there was any reference to it.
Dan of Dan’s Data said in a review of P4 heatsinks:
Intel have tried to convince people that bending the bejaysus out of their
motherboard is a good thing, because if something smacks the computer then
the CPU socket’s less likely to pop off the board… I remain unpersuaded…
These coolers don’t actually break anything, at least in the short term. They
just look as if they ought to.
At Intel, digging through the P4 support area, I came across this
document, which says:
…the thermal solution provides controlled compression on the processor
and socket. This compression results in curvature in the motherboard… It
is normal to observe a bow or bend in the motherboard. The level of bow or
bend depends on the motherboard material properties and component layout.
PDF document showed this photo, presumably to reassure (!?) the alarmed
What can be said? The pressure applied to the motherboard seems excessive.
Intel even cautions about making sure the distance between the back of the mobo
and the case is adequate to accommodate this bowing. Modifying the mounting
bracket to reduce the pressure would probably help motherboard longevity if
you plan to play around with fans, heatsinks, etc.
I did come up with a solution to reduce the bowing; more on that later.
Intel P4 HSF Shuffle
One month after the purchase of the above retail P4 package, I purchased a
second one, ostensibly the same package as the first. In early March, Intel
caused some excitement among hardware nuts when they released photos of a new
heatsink they said would soon be shipped with P4. The photo showed an AVC Sunflower
HSF, which had already been reviewed as an excellent performance HSF. So I had
hopes that my 2nd P4 might be so equipped.
Alas, no such luck. What I got instead was a plain-Jane rectangular block HSF
What gives? I don’t know. Intel’s attitude seems to be: They’ll take what we
give ’em. They don’t have a choice.
As mentioned earlier, I would try the retail HSF at stock voltage, and then
experiment with lower voltages, resorting to a different fan or HS only if necessary.
2. Power Supply
Aside from the heatsink fan, there are two other fans in these components, both of them in the SH ATX365P4 power supply, a 365W unit similar to the Enermax PSUs. The SH brand is associated with Superpower.com which also imports Landmark ATX cases into the US. A couple of emails sent to Superpower went unanswered, so I can only tell you what I learned first hand about this PSU.
I used a jumper to start the PSU without a motherboard in order to listen to it. The PSU fans are fairly quiet upon start-up. However, even without any load on the PSU, both fans accelerated to a much higher speed within 10 minutes and the noise emitted by the PSU became more considerable. Modest by normal PC standards, but unacceptable by the standards of Silent PC Review. This in a room where the temperature rarely exceeds 20? C.
I decided to modify the PSU before assembling the system by removing the second fan, repositioning the internal thermistor away from the hot coil to which it was glued, and replacing the first fan with my standard quiet fan, a Panaflo 80mm FBA08A12L. Note that opening up the case voided the manufacturer’s warranty on the PSU and exposed me to some danger from electrical shock. I worked with the AC cord unplugged, of course. (Please see my article about silencing PSUs with thermistor-controlled fans.) BTW, the Panaflo fan is probably the most popular quiet fan among silent computer builders and modders: rated at a low 21-dBA, 24 CFM airflow, and widely available for as low as $3 from some US web retailers.
When the modification was completed, the noise of the PSU dropped to the level of a whisper and never rose under any condition. Yet the soft airflow from the Panaflo fan in the PSU never got more than warm. It became inaudible inside the case when placed under my desk.
The other noise makers among the components were an older 4x Creative CD-RW
drive and a Seagate
Barracuda IV 20G hard drive. The slow CD-RW makes little noise and is of
consequence only when it is being accessed, so I would install it normally.
The Seagate Barracuda IV has been the quietest 3.5″ hard drive available ever
since it was introduced in the third quarter of 2001.
In my experience with a large variety of other 5400 and 7200 rpm IDE drives
from IBM, Maxtor, Quantum, Western Digital and even Seagate over the past several
years, nothing comes close to the single platter 20G and 40G Barracuda IV models
I began using late last year. With noise rated at a vanishing low 20 dBA while
idling, the most notable aspect of these 7200 rpm HD is the complete absence
of the high pitched whine that characterized all other 7200 rpm drives in my
experience to date.
Still this does not mean the Barracuda IV cannot be made even quieter.
I recently devised an elastic suspension system for hard drives. It ensures that the noise of Barracuda IV remains unchanged even when mounted in a case, which usually amplifies drive noise by resonating and vibrating in sympathy. This suspension would be used for the Seagate. (See this article on hard drive silencing.)
I usually opt to maximize ventilation in a solid, heavy case and add sound
damping to the interior. But this time, I used a cheap, flimsy generic mid-tower
(18″ high) ATX case for 3 reasons:
With a flimsy case, solidity and thickness went out the window. Improved ventilation
for better and quieter airflow was definitely achievable, however.
Out came the power drill, jigsaw and the metal cutting blade. The photos below
are self-explanatory. The back panel fan “grill” was cut away to improve airflow
and reduce air turbulence noise. The hole in the top panel was made as a “convection
outlet” for the PSU. I planned try to run the PSU without any fan, mounting
it upside down so that the hole for the second inside fan would be facing up.
More on this later.
The grill material for the front panel fan was also removed. I wasn’t sure
I would use a front fan, but the enlarged air passage hole would help airflow
and reduce turbulence. But cutting the metalwork away would do only half the
job, because the outside air-in duct is actually a small slot opening at the
bottom of the front plastic bezel. So the next step was to enlarge the opening
on the front bezel.
Cutting through plastic with a jigsaw is amusing. The plastic
cuts easily, but also begins to melt as the blade heats up. Then you have to
quickly bend the plastic away from the cut so it won’t bond back together as
Both fans were removed from the PSU before installation in the case. As mentioned
earlier, the PSU was mounted upside down. This configuration places the PCB
in the PSU at the bottom, and convection can now occur naturally from the unblocked
heatsinks. The cover was left off so that the air from within the case would
easily rise up around the sides of the PSU and exit through the hole at the
top or at the back of the PSU.
On the right photo above, note the green insulated wire being used to secure
the PSU in place. With the cover off, the PSU lacks the rigidity to be held
properly by just the 4 normal screws on the back panel. The CPU HSF is visible
just below it, and to the left, a Panaflo 80mm fan on the back panel, mounted
the wrong way here — it was meant to draw cool air into the hot area around
the CPU and the Northbridge chip, then rise up and our through the hole above
Later, I changed my mind about the effectiveness of the back panel Panaflow
and repositioned it over the video card and the Northbridge chip, as both appeared
to get quite hot. (See the photo below.) They benefited considerably from the
added cooling. I believe system stability was improved by this fan.
NOTE the metal piece used to suspend the fan over the chip and the video card. It’s a PCI slot cover that gets taken off the back of any case whenever you use a slot. Modified with a hole drilled at one end for the single screw that holds the fan. A little bending, and twisting, and then tightening of the screws is all that’s needed. My poor man’s Zalman-style fan bracket.
Below is a final shot before the front bezel was installed.
There were no problems installing everything into the case. Once all the components
were wired up and running, I made some basic adjustments to the BIOS, then did
a fresh install of Window 98 SE, along with the usual laundry list of updates
& fixes. After all this tedious work was done, the boot-up time of less than
20 seconds made me quite excited.
The AMI BIOS on the MSI 645 Ultra motherboard allows the system clock to be
changed in 1 MHz increments from 100 to 200 MHz. I systematically increased
the system clock by 5 MHz, starting at 100 MHz, running Sandra 2002 Burn-in
for just 10 cycles at each new speed as a quick check on stability.
One frustration I experienced with the motherboard is that the temperature
monitoring does not work. In Motherboard Monitor 5, the gauge stayed stuck at
50C. In the BIOS, the temperature reading began at 30C, but the system would
get unstable when it read only 40-42C, which is far too low. MSI’s own forums
are full of complaints about this problem.
The system actually booted and ran at 2.4 GHz (150 MHz clock), but not stably
nor long enough to get any benchmarks. At 2.32 GHz (145 MHz clock), it did a
little better but still remained a bit unstable. Backing off to 2.25 MHz (140
MHz clock) made it all more stable. At this clock speed, I increased the memory
timing from Normal to Fast, then Turbo. The system remained stable enough to
run some SiSoft Sandra 2000 benchmarks, screenshots of which are shown below.
CPU core voltage was set to 1.6V, which both the BIOS and Sandra reported as
1.55V. This is 0.05V higher than the default of 1.5V. I have seen reports in
forums that this motherboard does provided slightly low Vcore. Whether the CPU
speed limit is due to limitations in the the memory, the motherboard or the
CPU, I have no way of telling without at least trying different memory.
I offer no interpretation of the above, only a summary: This is a fast, powerful yet inexpensive system.
One sharp-eyed reader pointed out that the memory bandwidth benchmark results seem poor. I did get much better reading later, in the 1900 MB/s, close to 2000. These were early benchmark efforts with the memory clock set too low – probably at 200 MHz instead of the standard 266MHz.
No Game Benchmarks
I don’t wish to tell you how Quake or Unreal or whatever other
shooter games do on this machine, because I don’t play these games. The only
game I play from time to time is the real-time strategy game Red Alert 2
& its follow-up, Yuri’s Revenge. On my other systems, these games play
fast with excellent graphics & movement. But one area where things get bogged
down is in missions with very large maps, where the scroll function would sometimes
crawl. This slowdown does not happen the P4 system, and at the fast setting
in skirmish missions, the computer became far to fast for me to win.
My main work-related programs all benefitted somewhat in comparison to my main
PC, a 1-Ghz AMD T-Bird with similar components and PC-150 ADRAM. The Adobe programs
Photoshop, FramaeMaker, Acrobat and InDesign as well as AutoCad, Visio, and
Word all ran a little more smoothly. Some of Photoshop’s more specialized CPU-intensive
functions did seem dramatically improved.
If anyone wants to see benchmarks, there are enough reviews of the P4 Northwood
on the web that offer pages and pages of such info.
After the initial excitement of a 2+ GHz computer, I realized there were some
subtle misbehaviors that I could not explain. One example was with PDF files.
I build them using Acrobat. The build process would go fine, and the PDF file
would work perfectly fine on the system. But the PDF file would cause errors on my other
system, regardless of how it was accessed. I attributed this anomaly to the
overclocking, and backed the clock speed down to 2 GHz (125 mHz x 16) where
the system performed impeccably.
What About Noise?
When first assembled, the system had 3 sources of noise:
The PSU, as mentioned, was initially installed without any fans.
The system was fairly quiet, but far from silent, certainly noisier than my
reference systems. The original Intel fan was by far the noisiest component,
with a whine typical of smaller fans. Dropping the feed voltage to 5V by using
at IDE power connector made a huge difference.(See this article if you don’t
know how to get 5V for your fan.)
I then made a change that increased the noise: I added a Panaflo 80mm
fan at minimum speed to the SH power supply. Perhaps I was overly cautions;
I was not comfortable with the heat I could feel rising up from the PSU through
hole in the top of the case. Perhaps it was perfectly safe, but unlike silent PC pioneer Cub Lea I was
not ready to do this.
The Panaflo went in easily. The little tabs on the 2-pin connector that came
with the Panaflo had to be cut away, but then I was able to plug it into the
pin connectors on the fan PCB. The cover went back on, and the PSU mounted normally
with the opening for the 2nd fan over the CPU area.
With the thermistor in the PSU repositioned for minimal heat exposure, the
Panaflo fan gets just under 5V. Enough to start, but inaudble beyond about 2-3
feet (depending on ambient noise and your hearing). It reduced the temperature
of the exhaust air from the PSU from hot down to merely warm.
Before & After
I decided to record the before and after sounds
with the Windows sound recorder, using a SBLive! sound card and and a cheap
microphone. The mic was placed on my camera stand & positioned about 10-12″
away from the CPU. 44KHz 16-bit mono quality was selected. After the recodings were made, they were converted to MP3 format to reduce the file size, down to around 130kb from the orignal 1.3 to 1.5 mb.
Clicking on each of the 3 speaker / sound images below should enable the MP3 files to play. Ajust your speaker volume up whne playing the first file, then do not adjust them again in order to get a sense of relative loudness.
** Virtually silent reference system: P3-550 oc’d to 733 via FSB,
lowered VCore (1.45V), GlobalWin VOS32 heatsink w/o fan, generic PSU w/ fan
@ 4V, one back panel Panaflo 80mm fan @ 4V, Seagate Barracuda IV HD in novibe
style mount, generic 18″ ATX case, well damped w/ highest density carpet underlay
foam. This system is only audible in the wee hours of the morning when ambient
noise drops to tomb levels. It makes about as much noise as a single Panaflo
80mm fan running unobstructed at 5 volts or less.
Replacing the Stock Retail Fan
The noisiest component of the system at this point was the stock Intel fan.
Even at 5V, it could still be heard, with a bit of whining and clicking noises
from the bearings. Because the frame of the fan actually comprises a basic part
of the heatsink mounting system, replacing it without replacing the heatsink
meant that something had to physically take its place.
Here’s how I did it: I cut a small piece of wood the length and width as the fan assembly, and cut out the middle. (The wood is actually a piece of a tonearm board used on the famous Linn Sondek LP12, a fact that may amuse audio aficionados.) It is slightly thinner than the area of the fan structure used to clamp the assembly down. Thus, there is less pressure on the motherboard, and less bowing. The heatsink is still on very securely; I could not twist or move it in any way after it was clamped down under the piece of wood.
Another 80mm Panaflo fan was mounted over the CPU heatsink. It, too, runs at
5V off an IDE power connector. This is essentially inaudible.
This article posed the question: Can a Pentium 4 be made to run silently at 2-Ghz using non-exotic, inexpensive techniques? The answer is a surprising and resounding YES!
* * *
Discuss this article in our Forums.