Pentium M for the Desktop: AOpen i855GMEm-LFS & DFI 855GME-MGF

It is a tome about the merits of the Pentium M 2.0 Dothan with 855-chipset desktop boards by AOpen and DFI for silent, powerful computing. We’ve waited 18 months for these boards and our Pentium M desktop platform review was two months in the testing and writing.

February 8, 2005 by Ralf
Hutter

When Intel launched the Pentium M processor for notebooks, Silent PC enthusiasts clamored that it might be the perfect
CPU for a really quiet system. The extremely low power consumption and the ability
to dynamically undervolt and underclock depending on CPU load are the sort
of things that makes silent PC geeks salivate. However, it seemed like Intel
never intended the Pentium M to be used on the desktop, so all we’ve
done for the past 18 months is wait and dream.

Back in 2003, the Pentium M, with its sub-2GHz clock speed and short pipelines
would have been the odd man out if Intel had positioned it to compete
on the desktop with the P4 Northwood and it’s 20-stage pipeline.
How could Intel have justified its “higher Megahertz is better” marketing given the incredible performance of the lower speed,
high IPC Pentium M? Well, back then they didn’t have to, but along
came Prescott in 2004.

Intel’s move to the 90nm 3+GHz Prescott was a godsend
for aftermarket heatsink manufacturers and air conditioning companies. The Prescotts put out
over 100W and don’t really perform any better than their Northwood predecessors.
The increased cooling requirements and sub-par performance of the Prescott haven’t
earned many fans among gamers and other performance-oriented users, and
it certainly hasn’t been getting good press among quiet PC users.

The mass market seems relatively unaffected by the Prescott’s
shortcomings even as the enthusiast segment of the market gravitates to
AMD’s Athlon 64 as it currently takes top spot for the processor of choice due to its slightly higher performance and lower cooling demands. The latter, especially, is an odd reversal, given AMD’s hot chip reputation in the recent past.

There have been some early industrial boards for the Pentium M, but these were not equipped with AGP slots or consumer expected features. They also cost about four times as much as a typical P4 board. Finally, at the end of last year,
AOpen and DFI released
desktop boards for the Pentium M based on the Intel 855G chipset. These new boards give Intel
a new entry in the high performance desktop market, and give
the quiet PC crowd a chance to finally see if the Pentium M is all
that we’ve been hoping for.

2.0GHz of Pentium M (Dothan) goodness.

The AOpen and DFI boards differ
from earlier industrial Pentium M boards in that they have
features like an AGP slot, two DIMM slots, extra PCI slots,
onboard S-ATA and Firewire, adjustable BIOS and they’re
available for end-user purchase through normal retail channels. While neither
of these boards are inexpensive, they’re priced about 30-50%
less than the industrial boards.

SilentPCReview hustled and cajoled to obtain early samples of these new boards, along with a new Pentium M 2.0, Dothan core. They’ve been in my possession on the test bench and on my desktop since before Christmas. It’s been a long haul to this review.

The retail packaging for the P-M boards.

AOpen i855GMEm-LFS

Click for large image of AOpen i855GMEm-LFS.

AOpen’s entry is a full featured Micro-ATX
board that resembles their typical socket 478 motherboards, even down to the
heatsink retention bracket around the CPU socket and the black colored PCB.

The Intel mPGA-479 CPU socket is in the usual location near the
top edge. This socket looks similar to socket 478
for P4, but their pinouts are electrically
different, meaning processors will not interchange. The Socket
479, while still a ZIF design, holds the CPU in place with a 180° turn of a locking screw
instead of a lever. This low profile socket design
is most often found in notebooks and quite easy to use.

AOpen board showing low profile CPU socket and the handy P4-style retention
bracket.

The socket 478 plastic heatsink retention bracket is probably the nicest features of the AOpen board: It lets
you choose from an ocean of P4 heatsinks.
AOpen actually includes a CPU cooler, a simple finned aluminum
heatsink with an 80mm x 15mm fan. The fan is rated for 1900 rpm and <24dBA.

Included HSF cools well and is pretty quiet.

Next to the CPU socket is the 855MCH chipset. This “northbridge”
is passively cooled with an extruded aluminum heatsink, cleverly formed into
the name AOpen. Nothing like a little free advertising,
eh? [Editor’s Note: This AOpen logo HS is also used in some of their SFF motherboards.]

“Function follows form”? AOpen’s passive NB heatsink.

The 855 MCH supports up to 400 MHz front side bus speeds, pedestrian
in these days of 800 MHz and 1066 MHz FSB. The 855 chipset allows a maximum of 2GB DDR memory on two DIMM slots and only supports single
channel DDR-333 memory. This limits peak memory bandwidth to 2.7
GB/sec. In contrast, P4 boards with dual channel DDR2-533 have memory bandwidth levels of 8.6 GB/sec.

The ICH4 Southbridge
is a bit dated, as it was the common I/O controller hub for second generation
P4 processors on 845 chipsets. While the ICH4 supports
PCI, USB 2.0 and IDE storage, other features such as S-ATA, RAID,
PCI Express slots, or HD Audio are not found in this older chipset.
AOpen had to go with third party hardware to fill out the feature set. A Promise S-ATA II controller chip supports two S-ATA hard drives in standard RAID-0, RAID-1 or JBOD modes.
Because the chip is S-ATA II compatible, it also
supports Native Command Queuing and Staggered Spin-up.

ICH4 southbridge chip. Also note SATA-II, USB 2.0 and Firewire headers at
bottom of board.

The onboard audio supports basic AC’97 audio through a Realtek controller for 5.1 analog audio output, but no digital output. The
motherboard itself contains the pinout for a SPDIF digital connection but
there is no hardware for external connection to the ports.

The board also features built-in Firewire 400, but no external ports on the I/O panel or PCI slot hardware are provided. There is an
Agere Firewire 400 controller, which is output only
through two Firewire pinouts. This allows you to connect remote front or back panel Firewire ports. One nice feature is the colored
front I/O header, which makes it easy to hook up the front panel wiring
when the board is installed in the case.

Front I/O header is color coded.

The board’s AGP slots is only 4x but it can still run modern AGP cards like the
Nvidia 6800 and Radeon X800. Today’s fastest AGP cards rarely
reach the bandwidth limits of AGP 4x, so this is not much of a limitation. Also included are three 32-bit PCI slots, which is standard for a m-ATX board.

An AGP slot and some spare PCI slots!
This unique concept makes this board useful to the home user compared to early
industrial P-M boards.

Being the “G” version of the i855 chipset, the board also includes
Intel’s integrated Extreme Graphics 2 onboard video, suitable
for 2D applications but not up to snuff for 3D gaming or anything multimedia
intensive.

The board is equipped with two Marvell Gigabit LAN controllers. These
chips run off a 32-bit PCI bus, so they may not be as efficient as a CSA
controller, but they should be sufficient for most applications.

The ATX I/O panel includes the usual group of PS/2 mouse & keyboard connectors,
as well as a parallel and serial port, HD-15 VGA connector, the four USB 2.0
ports, dual LAN ports, and the three multifunction analog audio jacks. As noted
earlier, there is no external Firewire port or external SPDIF connector on the
I/O panel.

AOpen provides all the typical I/O connectors along with an extra GigE LAN
port.

The board is powered by a standard 20-pin ATX plus an auxiliary 4-pin 12V connector. Since
real-world power consumption of these Pentium M CPU’s is so low, power
supplies rated as low as 250W would be plenty for a system such as this, even with
a power-hungry GeForce 6800 vidcard installed.

Our board came with these accessories:

• low
  profile CPU cooler
• SATA data cable
• pre-folded flat ribbon IDE cable
• pre-folded floppy ribbon cable
• SATA power adapter
• ATX I/O shield
• Promise
  RAID drivers on a floppy
• CD with drivers and utilities
• well written
  user manual.

AOpen
doesn’t highlight its most interesting utility, the Series Tool,
which lets the user control the CPU FSB and multiplier as well as the speed of the CPU cooling fan from
within Windows. It also features a “SpeedStep” mode which works with the CPU’s
architecture to dynamically adjust the multiplier up or down in concert with
the CPU load .

There is a manual setting to
preset a series of speeds for the CPU to run depending on the load. The “Fan Control” setting allows one to set the fan speed to
full speed, a fixed speed that is some percentage less than full, or an automatic
mode based on temperature parameters set by the
user. The tab marked “Overclocking” surprisingly enough, let’s you
overclock the CPU by adjusting either the multiplier or the FSB. A “Hardware
Monitor” tab keeps track of the CPU and system temps, as well as the system
voltages. There’s little documentation about this tool on their website and
nothing at all in their manual.

DFI 855GME-MGF

Click for large image of DFI 855GME-MGF.

DFI’s entry into is also a full featured
Micro-ATX board, although its industrial roots are more evident.

The obvious difference between the boards is the small Northbridge-esque HSF that DFI provides for the CPU. CCertainly a 20W CPU won’t require much in the way of cooling horsepower, and DFI has chosen to embrace the “smaller is better” school of heatsink design with this board. The 2″ x 2″ square aluminum finned
heatsink appears to be lifted from a standard northbridge chip,
even down to the cheesy looking NMB 1604KL-04W-B49 40mm x 10mm cooling fan.

The heatsink is attached to
with four spring-loaded screws that bolt through the motherboard to a metal support on the back side. It’s an easy and
safe way to mount a HS on the unprotected Pentium M.

DFI heatsink sure is tiny. Retention bracket goes underneath board.

Underside of CPU heatsink shows its Northbridge lineage.

The CPU socket on the DFI board is the exact same low-profile version found
on the AOpen board. There are a couple of caps and some other PCB components close to the socket, which limits the size of any third-party heatsink
to replace the tiny stock cooler.

Low profile mPGA479 ZIF socket locks/unlocks with a 180° twist of the
slotted screw.

The DFI board runs the same 855G MCH chipset as the AOpen board and so includes
the same default features such as the 4x AGP slot, integrated Extreme Graphics
2, 400MHz FSB (although DFI recently released a new BIOS that gives 533MHz support),
and 2 x 333MHz DIMM slots. The chipset is passively cooled with an aluminum
heatsink that looks like a near twin of the processor heatsink, sans fan.

DFI’s NB heatsink looks less exotic than AOpen’s, but still gets the job
done.

Southbridge-wise, the DFI board uses the newer
Intel 6300ESB I/O Controller hub The 6300ESB is a more modern design that features
two Serial ATA ports (which can support RAID 0 and 1 configurations, as well
as JBOD), two IDE channels, four USB2.0 ports, two 33MHz PCI slots and a PCI-X
2.2. The PCI-X slot supports the most recent version of the PCI-X standard,
offering a 66MHz DDR 64 bit interface. The 6300ESB sports a black
anodized aluminum heatsink.

6300ESB Southbridge is cooled by this black anodized heatsink.

The board uses a standard 20-pin ATX power connector, and this DFI does not
require a dedicated 4-pin +2V ATX (“P4”) connector. This board will work with a lot of smaller or older power supplies.

The board’s I/O panel sports PS/2, serial, and parallel ports on the left along
with USB 2.0, Firewire, and Gigabit Ethernet connections in the middle. The
four USB 2.0 ports on the I/O panel are the only ones on the entire board; there
are no internal USB pins. This is a limitation of the 6300ESB chipset, and a bit inconvenient
in these days of USB flash drives, card readers, cup heaters, and other
USB gadgets. There are two internal Firewire headers which can be hooked up
to the front of the chassis.

The DFI’s ATX I/O panel sports all the usual connectors.

Over on the right side of the I/O panel are the audio connectors. The
board has a standard 6-channel analog audio setup connected through a Realtek
ALC655 AC ?97 chipset. DFI has not provided the external output connectors to take advantage
of the S/PDIF digital audio
outputs, although there are internal S/PDIF ports on the board.

Our sample was built on a yellow-brown PCB but production boards will be fabbed in sexier black. Accessories
include the HSF, two pairs each of SATA and flat
IDE cables, an SATA power adapter, rear ATX Shield, a floppy with
RAID drivers, extra jumpers, and a CD containing required
drivers and a manual in PDF format.

DFI on the left, AOpen on the right.

TEST SETUP

We will build a couple of test systems and run some
basic benchmarks and examine in detail the thermal performance of these
Pentium M systems. We will also experiment with various configurations to answer
some burning questions:

• Just how cool does a 21W CPU run?
• How
  much current does a Pentium M system draw under full load?
• How
  well do the default cooling systems work and are they quiet enough?
• Will the stock coolers work well enough to be run passively?
• Will
  third-party cooling solutions be more efficient than the default
  coolers?
• Will third-party cooling solutions allow the CPU to be passively cooled?

Since the Pentium M CPU runs so cool, we’re going to utilize the quietest hardware
available. If our goal is to build a fast and extremely quiet system, the
ancillary hardware shouldn’t drown out the CPU cooler. Our test system will be based on the following hardware, all of which is
ridiculously quiet:

TEST PLATFORM

The test platform is an open system not enclosed in a case.

Intel Pentium M 755 – 2.0GHz Dothan core – TDP is 21W, MP
is 26.8W.
Sapphire ATI Radeon 9600Pro passively cooled video card (AGP)
Mushkin PC3200 Level II – 2 x 512MB DDRAM @ 2-2-2-5, 333MHz
Samsung MP0402H 2.5″ 5400rpm, 8MB cache notebook
hard drive, decoupled with Sorbothane blocks (<17 dBA/1m)
Seasonic Super Silencer 300W (rev A1) PSU modded with 5V Panaflo M1A
(<18 dBA/1m)
Arctic Silver Ceramique Thermal Compound
Two-level metal platform with rubber damping feet. Motherboard on top; other
components below.
CPUBurn processor stress software
Motherboard Monitor 5.3.7.0 software to track CPU temperature
and fan speed
Seasonic Power
Angel power monitor used to measure system power usage

Pentium M test platform. Note the 2.5″ notebook drive on top of the FDD.

Each motherboard was tested using the same components.
Only the board and the various heatsinks
changed.

Due to the DFI CPU cooling setup,
we were not able to use typical aftermarket CPU cooling hardware. Much
of the DFI testing was done with the stock heatsink. Prior to all testing, the heatsink was installed
as per the manufacturer’s and Arctic Silver’s instructions.
No benchmarks or thermal tests were run until system stability was assured by
running the Prime95 Torture Test
(v23.8) for 8 hours and Memtest86 (v3.20)
for at least 24 hours.

Each thermal test was run for
30 minutes even though temperatures generally stabilized within 15 to 20
minutes. Each test was repeated three times on consecutive mornings to check
consistency. All results were within 1-2°C of each
other and the average readings are included in the charts.

No tests were run unless the ambient temperature was at the reference
level of 71°F (21°C)

* All temperatures in degrees Celsius.
* Diode: Reading from Pentium M 755 CPU diode via Motherboard Monitor.
Diode was calibrated on each board using the standard SPCR
CPU Diode Calibration method.
* Temp Rise refers to the difference between ambient temperature and
the diode reading. .
* °C/W refers to the °C rise per watt of heat
dissipated by the CPU.

SYSTEM ASSEMBLY

The
systems were easily assembled as each motherboard uses standard components laid
out in the usual “ATX” style. The low profile mPGA479 ZIF socket was a pleasure to use. I found
the camming action of rotating a little screw 180° easier than using the typical locking lever.

The Pentium M is the first CPU without an integrated heat spreader that I’ve used in years. So
it was with some trepidation that I installed the AOpen
heatsink for the first time. I flashed back to my AMD Thunderbird days and listened
carefully for that stomach churning cracking sound as I put pressure
on the heatsink, but everything went just fine. The DFI heatsink was even easier
to mount, due to the no-fuss springloaded screws.

This comment is relevant mostly for reviewers, but… One nice thing about the small, exposed core of the Pentium M: It takes just a tiny bit of thermal interface goop to cover
the CPU core, which is also faster and easier to clean
compared to an integrated heatspreader. This was a consideration during the extended testing of these boards. It seemed like I was
swapping the motherboards every time the wind changed directions.

Both builds went perfectly with each system reaching POST on the
first boot. I flashed the BIOS of each board to the latest revision and then installed Windows
XP sp2. After Windows was installed I loaded the latest Intel chipset drivers
and then installed the NIC drivers from the supplied driver disk. Once connected to the web, all the Microsoft OS updates were installed. Then, a minor amount of GUI configuration, and on to the benchmarks.

Each system in stock configuration. AOpen on top, DFI on bottom. Notice difference
in heatsink size.

SYSTEM BENCHMARKS

Running system benchmarks is
not SPCR’s raison d’etre, nor even our normal practice. Still, we wanted
to get an idea of the capabilities of this processor and chipset
combination, and to compare the two boards. All benchmarks
were run with each system in default configuration. FSB and multiplier
were set at the stock speed of 100 x 20. Vcore was left at default, which
is 1.324V. Each board was run with the same set of peripherals, except for the CPU coolers, running at 12V.

We also ran some of these benchmarks on a standard P4 system.
All hardware was the same except for an Asus P4P800-D motherboard, a 3.0C Northwood
P4 CPU and a Thermalright SLK900 heatsink. The same vidcard, memory and HDD
was used.

Sisoft Sandra Benchmarks

We used the full
registered version 2005.1.10.37. Benches run were the Math, Multimedia and Memory
Bandwidth. All the tests were run at default settings. Math and multimedia benches show
the 2.0GHz Pentium M holding its own against a 3.0GHz Northwood. Memory performance
is lacking, but that’s not unexpected since the Pentium M memory is running
in single channel mode at a slower speed than the memory on the P4 board.

Memory benches. Pentium M run with single channel, 333MHz RAM, P4 with dual
channel, 400MHz RAM.

Math benches. 2.0GHz Pentium M is holding its own with 3.0GHz P4.

Sandra Multimedia benches.

Next we ran some FutureMark benchmarks. We used registered versions of
3DMark 2003 (v.3.50, patched), 3DMark 2005 (v1.10), and PCMark 2004 (v.1.20).
All tests were run at default settings.

We also ran the PC Mag / ZDNet series of benchmarks, the Business Winstone
2004 (v1.01), 3D WinBench 2000 (v.1.1), and CD WinBench 99 (v.3.0). All tests
were run with their settings at default. The Pentium M boards performed almost
identically.

The Pentium M boards perform almost identically,
and give a 3.0GHz P4 system a pretty good run for the money. Each board was also used as my day-to-day system for about
four weeks. For the most part, the P-M seamlessly replaced my normal 3.0Ghz P4
system. Subjectively, the only slowness that I noted was on HDD-intensive tasks,
probably due to the 5400 rpm notebook drive
used. Overall, the performance of this 2.0GHz Pentium M was very impressive.

SYSTEM OPTIMIZATION

A. Thermal Testing and Noise Analysis

Each system was first tested with its
default heatsink and fan, with the fan running at 12V and the CPU running at the default
Vcore. The little
40x10mm DFI fan at 12V had that annoying, high-pitched whine that is common with
most small fans. There was also a fair amount of air turbulence noise. It wasn’t
as bad as many small fans that I’ve heard, but it was definitely too loud to live with.
The 80x15mm AOpen fan, on the other hand, wasn’t too bad at 12V. It sounded more like
a decent 92mm fan. It had a sort of low-pitched mechanical
whine and a bit of air turbulence noise. Overall temperature rise under load was pretty
low.

Next I tested each heatsink with the fans running at 5V. The DFI fan lost
a significant amount of its whine at 5V but there was a noticeable clicking
noise now present. And it obviously wasn’t putting out much airflow as
thermal performance became awful. The AOpen fan was surprisingly quiet,
and basically silent from more than 2′ away, but a bit of metallic clicking
noise was more evident now that air turbulence noise was reduced. Performance was adequate, especially given the low noise. I could live with this fan at 5V, and it might
be the perfect HSF to use in tight spaces (think HTPC or SFF) with its low overall height of only 1½”.

B. Thermals and Noise with Reduced Vcore

Both motherboards have user-friendly BIOSes with lots of adjustable
parameters. In particular,
CPU voltage is adjustable from the default setting of 1.324V down
to a minimum of 0.875V.

Overclockers won’t be so happy as neither
board allows any Vcore overvolting . This may not be a big issue,
as other web sites have reported 20-30% overclocks of their Dothan P-M processors at stock Vcore. We tried some overclocking with each board: Both were able to run ~ 20% overclocked at the default Vcore. The
DFI board is a little more OC-friendly as it has a larger selection
of dividers to keep the memory and PCI/AGP bus speeds under control when the FSB
of the CPU is increased.

Lowering the voltage
of the CPU while keeping the CPU speed at default reduces CPU heat output greatly without any performance penalty. Less heat equals less need for cooling, which allows for slower,
quieter fans, or maybe even no fan at all!

So it was with great anticipation that I approached this part of the testing.
I ran a series of tests to determine the minimum stable CPU Vcore. This was accomplished by
lowering the default Vcore in 0.1V increments, then running Prime95 for about
an hour. If the system passed, I lowered the Vcore by
another 0.1V and repeated the process. I kept dropping the Vcore till errors started started showing up in Prime95.
At this point, I increased the Vcore in small increments until Prime95 would run for 48 hours with no errors. Both boards gave me the same, 1.1V Vcore minimum
threshold for this particular 2.0GHz Dothan.

With the AOpen, CPUBurn load temp dropped by 11°C! That’s an 11°C temperature
drop with no loss of performance and no increase in fan speed. It’s like getting
a cooler CPU for free, and will actually save money due to lower power requirements
over the life of the CPU. The
DFI board got almost the same temperature reduction as the AOpen board,
an impressive 10°C.

Lowering the CPU fan speeds to 5V was the next step,
and these results were even more impressive. At the lower Vcore, the AOpen
load temp dropped from 54°C to 39°C, a 15°C drop. With the tiny DFI heatsink, it went from 73°C to just 45°C! That’s a 28°C
drop, just by lowering the Vcore by 0.22V!

So how much of a power drop effected such thermal reductions? To estimate power
dissipation of the Pentium M 2.0 Dothan at the reduced 1.1V, we used the nifty
utility CPU Power (v2.1) by Kostik. Simply edit the text file database
in the utility to add the data for this processor, then calculate the new TDP
value, which turns put to be 14.1W. It’s only a 6W difference, but it’s
also a 30% drop. This is roughly confirmed by our Seasonic
Power Angel power monitor which showed system AC power draw under full
load to be 56W at default Vcore and 48W when running the CPU at 1.1V. Factor
in the ~25% AC-to-DC conversion power loss and we have a 6W difference in DC
power as calculated.

Fanless with Stock HS?

The reduced Vcore testing showed amazing temperature reductions
with absolutely no performance penalty with both stock heatsinks,
neither of which is anywhere near “high performance”. Both stock fans are pretty quiet at 5V, but the
obvious next question is, “Can either of these heatsinks be run without
any fan at all?” My experience with the similarly cool
PIII Tualatin processors told me that they would probably run too hot, but since these
Pentium M’s are so well protected thermally, I decided to try anyway.

The fans were removed from both heatsinks, and the CPU temp
warnings in Motherboard Monitor turned off before running load tests. The test platform sits on an open bench, so there was no secondary
airflow from either case fans or a PSU fan.

On the DFI board with the tiny little heatsink, the CPU temperature slowly increased until it was at 75°C,
whereupon the system crashed. Motherboard temperature was barely 40°C at
this time. The CPU heatsink was too hot to touch for more than an instant. The fanless AOpen heatsink ended up in a BSOD at 76°C, although
it took longer to reach that temperature. CPU throttling
was monitored during this testing using Throttlewatch.
No throttling took place, even at 75-76°C. System temps as reported by MBM5, never got any warmer than mid/high 30°C
on either board during this fanless load testing. So, no fanless here,
at least not with the stock heatsink.

SILENT COOLING MODS

The stock heatsinks supplied with the DFI and AOpen boards are
good performers with their fans at full speed. They are probably perfectly adequate coolers with their fans
running at a much quieter 5V, providing the CPU is undervolted. But "adequate"
and "much quieter" aren’t what we shoot for here at SPCR.
After all, this is SilentPCReview, not sort-of-SilentPCReview. And these are the long-anticipated super cool yet fast Pentium M boards. So here is an assessment of a couple of modded, after-market, ultra-quiet cooling solutions for these P-M boards.

AOpen Socket 478 HS Options

With the AOpen AOpen
i855GMEm-LF socket 478 heatsink retention bracket, what kind of performance could an aftermarket, high performance
heatsink give us? On the other hand, perhaps the first question that must be asked is whether
we can even use a regular P4 heatsink on this Pentium M CPU?

The Pentium
M CPU socket has a lower profile, and the CPU itself is thinner due to the absence of the P4’s
Integrated Heatspreader. I broke out my fine measuring tools — calipers and depth mics — to check
the dimensions of the P4 and Pentium M heatsinks, CPU’s and sockets.
The overall height of the P-M mounted in its ZIF socket turns out to about .075-.078" shorter than a P4 mounted on a socket 478. This would indicate that any regular
P4 heatsink would need to be spaced down by about .075" on the AOpen.

But hold on: Intel’s P4
Thermal Design Guide calls for 75lb/ft ± 15lb/ft of clamping
force on the CPU heatspreader while the Pentium
M datasheet calls for a maximum of 689KPa, or 14.4lb/ft. Could this be due to the fragile,
unprotected die of the Pentium M? Possibly, but for whatever reason, the
clamping force on the P-M CPU is just one-fifth that of the P4. Lower CPU core height with a P4 HS would result in reduced pressure. Hmm… Perhaps the missing 0.075" on the Pentium M stack height is serendipitous!

I decided to try the trusty ‘ol Zalman
7000AlCu. The Zalman’s screw-down
mounting method would allow me to have a good feel for how much pressure I apply to
the Pentium M die. The 7000ALCu
was one that I had specially
modified in anticipation of something like this project. I swapped
the stock Zalman 92mm fan for one of the ultra silent (15dBA
@ 5V – mp3 sample) Nexus "Real Silent" 92mm fans. The Zalman
7000AlCu is a very efficient heatsink, but its fan is far from the ultimate in low noise. The Nexus 92mm is by far
the quietest 92mm fan I’ve heard, although it doesn’t blow tons of air.
I figured it would do OK on the 7000 in cooling a
21W CPU.
The Nexus fan + Zalman HS combo is extremely quiet; at 12V it is quieter than the stock Zalman fan running at 5V! At 5V, I can hear no noise at all from the Zalman / Nexus combo unless I position my ear just
3-4" away.

Zalman 7000AlCu HS modded with Nexus 92 fan on AOpen P-M board.

I placed the heatsink on top
of the CPU and started tightening the two Zalman mounting screws, paying special
attention to the force required. As the screws were getting close to bottoming
out, I could feel that the tension on the CPU was definitely less than if it
had been a P4. It seemed to have plenty of tension when the screws were all the way in. I wasn’t able to rock the heatsink, although I could
twist it a bit. I decided to fire it up and watch the temps.
If something didn’t seem right, I’d make some shims
to place on top of the Zalman mounting arms to apply more pressure to the CPU.
The idle temp stayed stable at only a few degrees
over ambient, and didn’t change with extra downward pressure on the heatsink. I declared the mounting a success and proceeded with thermal testing.

This HSF combo
does a fantastic job of cooling the P-M. There is basically no performance
difference between the fan at 12V or at 5V. SO… I tried the heatsink
without the fan. Amazingly, the temperature topped out at only 53°C
at full load, no matter how long CPUBurn ran. Fanless operation
with a Zalman 7000AlCu heatsink is a very viable option. Add
a bit of case airflow, or perhaps some ducting, and it will run even cooler.

On further thought,
I wouldn’t bother running the AOpen / Pentium M fanless. The 7000 with the Nexus at 5V is effectively
silent, and keeps the CPU over 20°C cooler than without a fan. (The fan is only flowing
a whopping 7 CFM and it keeps the CPU at just 10°C over ambient!) Having a bit of air blowing on the motherboard power supply components can only help. Motherboard makers (Intel included) rely on airflow from a HS fan to help cool the power components around the CPU.

Given that this Zalman is only one of a number of very high performance heatsinks, there are obviously many other options for fanless or low airflow cooling. All you need to look for is a high performance HS with good spacing between fins, the right orientation for good convection or auxiliary airflow from other fans in the case, and a good mounting clip that allows you to apply pressure safely or is easily adjusted for tension.

Alternate DFI Cooling Options

What about the DFI 855GME-MGF board?
The
hole pattern on the DFI cooler measures 40mm on center and 58mm diagonally which
is a bit different than the hole pattern on any NB cooler in our lab. Both Swiftech
and Thermalright
make high performance NB coolers and they have a decent amount of adjustability
built into the mounting system, which makes them usable… but they take only 40mm fans, so we’re not really getting anywhere here.

How
about a large fan suspended over the heatsink on something like a Zalman fan bracket or similar? After looking at both the Swiftech MCX159
and the Thermalright NB-1C, I opted for the Thermalright. It
has fins spaced wider apart than the Swiftech; it should work better with a low
airflow fan.

DFI with Thermalright NB-1C and 120mm, 5V Nexus fan.

The chosen fan was a 120mm Nexus "Real Silent" fan, perhaps
the quietest 120mm fan available. Due to its large footprint, it gives
some additional cooling to the Northbridge and memory DIMMs. It was
mounted above the board using a Sunbeam
Wherever PCI Rack. Testing commenced on the stock DFI heatsink with its
fan removed and the 120mm Nexus running at 5V. After this, the DFI heatsink
was removed and replaced with the Thermalright NB-1C and the thermal testing was
repeated. Both heatsinks cooled sufficiently in this manner, but neither did
a fantastic job. Low 50°C is probably perfectly safe, and the Nexus is nearly
silent at 5V so this could be an acceptable solution for cooling the DFI board
as quietly as possible. [Editor’s Note: Not all Nexus 120 fans start consistently at 5V; 5.5V is probably a safer voltage.]

This final round of testing proved that the AOpen Pentium M board
can easily be run with silent cooling for the CPU, whether from a 15 dBA/1m Nexus
fan modded Zalman 7000 HS, or a fanless Zalman 7000. The DFI is a bit more problematic due
to the tiny proprietary heatsink and non-standard mounting holes. Totally fanless is not an option, but the suspended 120mm Nexus does a sufficient job at a pretty darn
quiet setting. Modded as described above, either setup would
be quiet enough for even the most hypersensitive quiet PC enthusiast.

DYNAMIC CLOCKING AND SOFTWARE CONTROL

Dynamic speed and voltage adjustments based on work load is fully
implemented in Pentium M notebooks.Called SpeedStep, it works very well, and would be beneficial
for desktop users interested in running their system as cool and quietly as possible, much like AMD A64’s Cool ‘n’ Quiet.

Apparently, AOpen and DFI aren’t so sold on the SpeedStep
as neither company provided a fully functional version of this software.
Buried on AOPen’s Utilities Download page for the 855GME is a utility called
the Series Tool. This undocumented software contains AOpen’s version
of a SpeedStep utility. It also includes some rudimentary,
Windows-based fan control and overclocking software, and a system monitor
for temps and voltages. The software works, but it is pretty unstable and not really “ready for prime time”. DFI has no SpeedStep type of utility that I can find.

We played around with a few
experimental third-party, Windows based clockspeed utilities. RightMark CPU Clock Utility (RMClock)
essentially replicates the Intel SpeedStep software.
It functions on both motherboards, but like many other software-based clockspeed
utilities, it’s not very stable. It gives a tantalizing
taste of the power management possibilities on the Pentium
M platform. Here’s hoping the motherboard makers will release desktop software that works as well as the notebook
versions.

Eenie meenie minie mo

Both of these motherboards would be a fine choice for a quiet Pentium
M system. Is one better than the other? In terms of performance and stability,
they’re both excellent. The AOpen board is a bit more versatile, due to its standard P4 heatsink mounting bracket.
The DFI is a bit handicapped with its oddball heatsink, but it has better
BIOS options for the overclocker.

If I were forced to choose one board, it would be the AOpen by
a narrow margin, with the AOpen’s P4 heatsink bracket being the tie-breaker.
If the DFI board came with a P4 style heatsink mount, the nod would probably
go to it, based on its slightly better BIOS settings.

CONCLUSIONS

• Is it viable for
  day-to-day use?
• Is it an improvement over existing technologies?
• Is it the
  “be-all, end-all” for silent computing?

Having lived with these Pentium M boards and the 2.0GHz
Pentium M Dothan chip for the past eight weeks, I’m very impressed and will continue to
use it as my day-to-day system. The Pentium M CPU seems subjectively just
as fast as my 2.4GHz and 3.0GHz Pentium 4 systems, truly is silent from
more than 8″ away and only uses 40-50W during normal use. Each board
has been 100% stable and very easy to work with. Both boards sport the normal
complement of I/O, including SATA, onboard LAN, USB 2.0 and Firewire. Each
board is micro-ATX form factor, so they can easily be integrated into HTPC,
SFF and other small “Digital Home” computing setups.

Both boards could benefit tremendously from fully functional, stable
SpeedStep software.
In notebooks, SpeedStep extends battery life and helps the system run cooler; both traits
are certainly desirable in the silent desktop. SpeedStep can operate seamlessly in the
background, with the CPU running at very low power during most normal
computing activities, yet it instantly ramps up when CPU usage calls for it.
Even at full speed, ithe P-M 2.0 only puts out 21W, but stepped down to 600MHz, it
runs at a paltry 7W.

Is the Pentium M ground breaking? No, not really.

Yes, it’s
a powerful, low power consumption CPU, but with a little tweaking, the AMD A64
platform can cover similar ground, and costs less money. Some of us at SPCR have
been begging, pleading and cajoling Intel to release a Pentium M desktop board
for over a year now, and Intel still has not produced one. A year ago these P-M boards would
have been compelling, but now the A64 with Cool ‘n’ Quiet is a serious lower cost alternative.
Still, these AOpen and DFI boards have shown themselves
to be 100% bulletproof. Perhaps for those who value the “legendary
stability” of the Intel platform, the Pentium M may be the silent platform of choice. Others may vote with their wallets.

The Pentium
M systems are the the quietest system I’ve built, yet
they’re powerful enough for me to use as my main system. My test bench puts the
system 2-3 feet away at ear level, and with this platform I don’t hear any noise. The CPU isn’t
100% responsible for this as I’m using a passively cooled video card and an
ultra-silent Samsung notebook HDD, but the sum really is greater than the
parts. Building a quiet system starts with choosing the right parts, and I
don’t think you can go wrong with the Pentium M and either of these
motherboards as the foundation.

As it stands today, if money is no object to assemble the quietest PC possible, a
Pentium M based system is my first recommendation.

* * *

Much thanks to AOpen
and DFI
for the opportunity to review their respective motherboards.
Many thanks also to our buddies at Newegg
for the loan of the Pentium
M 755 processor that was used in this review!
And on the software side, thanks to Futuremark
and Sisoftware for the opportunity
to use their benchmarking applications.

And finally, thanks to Mom and Dad, I couldn’t have done it
without you.

* * *

Discuss this this article in the SPCR Forums.