Chris Thomson returns to SPCR with Round Two of his Pentium D silencing saga. As before, Chris does a great job with systematic documentation, and this time, he incorporates the feedback on his original DIY article from many forum members. By identifying names, their comments, and the way these comments were used for further improvements, the article also becomes a showcase of the SPCR spirit: A community sharing in exploration and discovery.
April 19, 2006 by Chris Thomson
Chris Thomson returns with Round Two of his Pentium D silencing saga. As in Quiet DIY OC’ed Pentium D 830 System, Part One, Chris does a great job with systematic documentation, and this time, he incorporates the feedback on his original DIY article from many forum members. By identifying names, their comments, and the way these comments were used for further improvements, the article also becomes a showcase of the SPCR spirit: A community sharing in exploration and discovery.Readers who haven’t read part one are encouraged to do so before plunging into the hot waters below. It’ll be fun to go through the forum discussion, too, and then see what Chris has done with some of the suggestions. Mike Chin, Editor |
I first wrote about this system in Quiet
DIY OC’ed Pentium D 830 System, Part One. That article ended
with the system
seriously overclocked, the case open, and cooling provided by
four AcoustiFan
DustProof fans running at 675, 685, 695 and 995 RPM, and aided by a
couple
of ducts.
Then the SPCR forum crowd weighed in. Brutal, but helpful. The
comments in the forum launched me on another round of mods. The goal was to run with the case closed, preferably with even
slower fans.
OVERCLOCKING OVERDONE
Before I started on those changes, I had issues with
system stability.
I had been running the system with the FSB/DRAM at 240/640
MHz, the GPU/GRAM at 427/1110 MHz, and the DRAM/CPU/MCH at
2.00/1.325/1.60 Volts. It could run Prime95, CPUBurn, protein folding
or PCMark/3DMark indefinitely, but had occasional odd hangs or slowness
running HD- or CD-intensive tasks. The behavior was consistent with
lost IDE interrupts. I tried dialing back the clocks a bit and these
symptoms disappeared. This led me to conclude that the performance
limit in my system is set by the chipset (north and south bridges), not
by the CPU or memory.
After some experiments, I settled on FSB/DRAM and GPU/GRAM
speeds of 235/627 and 427/1110 MHz. At these speeds, I was able to
reduce the DRAM/CPU/MCH voltages to 2.00/1.30/1.55. The CPU voltage
and clock reductions dropped its power consumption by 6%,
which made the VRM run cooler. CPU-Z reports the CPU clock as 3.532 GHz.
BETTER VRM AIRFLOW WITH A
BAFFLE
Rereading my article and looking at the initial responses led
me to the question: How can I increase the airflow over the VRM?
My
earlier thinking had fruitlessly focused on upstream
deflectors. What if I added a downstream baffle that would force more
air across the VRM?
This led me to the idea of a baffle that would force all the
top case fan air flow over the VRM parts by sealing off the other
paths.
A simple right-angle baffle that mated with the Ninja duct would do
this: the top flap of the Ninja duct ends 2 inches from the
motherboard, so a baffle that mated with the top and front edges of the
Ninja duct would force all airflow to the top fan to come from the VRM
region of the motherboard. Here is what the baffle looked like before
installation
Simple
VRM baffle, before installation (front view).
and after installation
VRM
baffle installed in the system.
The baffle is held in place by a couple of cuts into the near
top and near back foam, and by the CPU fan. It sits snugly on top of
the Ninja duct. To reduce noise, I added some 5mm foam to the inside
edges of the baffle, as well as the top flap of the Ninja duct.
This worked quite well: the exit air temperature at the far
edge of the top case fan dropped 1°C, and the VRM
components seemed to be a lot cooler, although they
were still very hot.
EXTRA VRM HEATSINKS
This led to the next idea from the SPCR forum, suggested by diver and golemB: Why not add
some copper to the VRM to catch the existing airflow?
I
bought some Swiftech
MC14 BGA RAMsinks and stuck them on the various hot spots of
the VRM. These are forged copper heatsinks designed for video card DRAM
chips, and seemed ideal for this application. Swiftech claims a
6°C advantage over the competitor. Since the PLL chip also
ran very hot, I added a heat sink to it as well.
To install these heatsinks, I had to temporarily remove the
Ninja. Here are before and after pictures. Note the discoloration of
the board and torroids near the VRM (top center) after running the
system for seven months with poor VRM cooling.
VRM and
PLL before adding heat sinks.
VRM and
PLL with heat sinks installed (and Arctic Silver 5 ready for Ninja
remounting).
Here is a photo looking down through the top case fan. The
copper at the top left is one of the VRM heat sinks. The whitish part
at the top is the base of the Ninja. The foam on the top of the Ninja
duct is also visible. The airflow to this fan is constrained to the
two-inch channel at the top of this photo. As described in Part One,
the foam and
coins in the center are to damp the ringing of the fan hub.
The combination of baffle and heat sinks dropped the VRM temperatures a lot. The main MOSFET heat sink went from “YOW!” hot to too hot to hold a finger on (subjective I know, but I think that would indicate at least a 10°C drop), and the heat sinks on the other MOSFETs and the PLL were now somewhere between warm and hot to the touch. I guess I should have measured the temperatures, but didn’t think to do so.
ADDING A VIDEO
CARD DUCT
SPCR forum responses from diver
and scara
asked: Why not move the video card
heat directly out of the box?
I thought about this for a
while, and came up with a duct that would capture some of the
motherboard fan air flow and direct it across the heat pipe radiators
and out of the box. The duct is composed of the video
card PCB as its bottom, the motherboard as its far side, and styrene
panels
as its top and near sides. The top of this duct would collide with the
north
bridge duct, so it requires a curved cut to let the two ducts mate
properly. Here is a photo of the video card duct before installation.
The notch at the bottom is for a capacitor on the motherboard, the
notch at the top is for the heat pipes, and the
deflector at the top is intended to smooth the air flow.
Video
card duct before installation, back/bottom view.
When installed, this duct sits on top of the video card PCB, and butts
against the motherboard and the back of the case, as shown in this
photo. The motherboard fan is just to the right of this photo. The rear
panel
PCI blank above the video card slot was removed to let the hot air out.
Video card duct after installation.
This duct removed 30~50W of heat from the motherboard air flow, allowing both lower temperatures and slower fan speeds.
OPENING UP THE UNUSED 5.25″ BAYS FOR BETTER AIR FLOW
An SPCR forum response from PhillyB
mentioned the Scythe
Kama Bay. This sparked the thought: Why not have
two ingress air streams, one feeding the CPU fan, and the other feeding
the motherboard fan?
If these worked reasonably well, it would be
possible to close up the case, and have dust filters on all air
streams.
The P180 already has a dust filter on the ingress path to the lower
half of the upper
bay, and installing a Kama Bay would provide one for the upper half.
A simple baffle could keep the two streams separated. This would double
the input air flow cross-section and nearly
double the total upper chamber air flow. The Kama Bay (with its dust
filter but without its fan) would occupy the three unused 5.25″ bays at
the top front of the case.
Here is a photo with all the fans, ducts and baffles
installed. The
Ninja, north bridge and video card ducts dovetail in the center, the
VRM baffle is above the Ninja duct, and the chamber baffle is wedged
between the motherboard fan and the DVD drive PATA cable. This photo
also shows the last two RAMsinks, mounted on the south bridge and the
TV card processor.
Two
fans, three
ducts, two baffles, and some copper.
The top-half air flow in this arrangement is straight through
the Kama Bay dust filter, the CPU fan and Ninja duct, then out the back
of the case.
The bottom-half air flow is more convoluted. It enters
through the P180 dust filter, goes through the motherboard fan, and is
split into three parts. The center part flows through the video card
duct and out the back of the case. The top part flows through the north
bridge duct, across the motherboard and VRM, into the top case fan
chamber, then through the top case fan and out of the case.
Finally, the bottom part swirls
around the south bridge and TV card, then flows over the near side of
the video card duct, then to the motherboard, where it joins the top
part flowing over the motherboard and out the top of the case.
Installing the Kama Bay required fitting the steel P180 drive
clips to its shell, which meant drilling some holes. These clips are
made from a very hard steel. Unless waste material is clamped to both
sides, drilling the holes causes even Bosch
titanium bits to wander severely. The plastic holders the
clips come on work well as bit guides. Remember: measure twice, drill
once! Here is a photo of the Kama Bay with all the hardware installed,
ready to be put into the case.
Scythe Kama Bay with P180 clips
attached.
The clips need to be bent
outward slightly to engage the chassis slots because the Kama Bay is a
bit
narrower than a standard CD/DVD drive.
Doubling the air flow in this
manner let me close the case and still use slow fans. The motherboard
temperature rose significantly (about 8°C) but not enough to
cause errors.
NOTCHING THE SLIDING CHAMBER
SEPARATOR
SPCR member BigA
suggested: Cut a notch in the rectangular sliding plastic that
separates the upper and lower chambers of the P180.
This allows the
SATA HDD data and power cables to go from the upper to the lower
chamber while still sealing off the air flow. Here’s a photo:
Notch in the upper/lower chamber
sliding panel to accommodate cables.
This was probably the
simplest mod so far, and worked well.
SWITCHING TO NEXUS FANS
At this point, I was running with four DustProof AcoustiFans,
all at
5V, their lowest reliable operating point. The system was very quiet —
the house and neighborhood had to be silent to have any chance of
hearing the computer noise. It was too quiet to record with the gear I
had. Interestingly, I could not detect any difference in sound level
with the case open or closed.
However, although the volume
of the noise was pleasingly low, the nature
of it was rather annoying. The CPU fan in particular put out a
thrumming sound reminiscent of a distant propeller airplane. The
motherboard fan also did this, but to a lesser extent. The cause of
this
thrumming was obviously the close proximity of the
downwind side of the fan blades to the Ninja fins and the motherboard
and
video card duct edges. Also, there was a distinct 245-Hz tone.
I could live with the thrumming per se, but
unfortunately
these two
fans were so close in speed that the thrumming beat about once a
second, making it very hard to ignore.
Thinking that a different brand of fan would run at a
different speed and get rid of the beating, I ordered a couple of Nexus 120mm
orange fans. These have long been MikeC‘s favorites,
and MadShrimp’s
comparison notes that the Nexus at 7V significantly
outperforms the AcoustiFan DustProof at 5V at about the same sound
level. I guess the reason I had taken so long to try these fans is that
I just don’t like the color. Why can’t they make some plain black ones
like everyone else?
When these fans arrived, I did some direct A/B comparison
tests. In free
air, the brand-new Nexus was slightly quieter than the seven-month old
DustProof at each voltage. Of course the Nexus spins slower at each
voltage, so that was no surprise. What was a surprise was the
difference in sound when I put a finger near the fan blades. Both were
annoyingly loud with an obstruction on the upwind side (this effect is
well documented on SPCR and elsewhere). However, on the downwind side,
the difference was striking: the Nexus was much quieter. This
difference appears to be due to the shape of the frame bevel, which is
deeper on the Nexus.
When I mounted a Nexus on the Ninja, there was no audible
thrumming anymore, and the 245-Hz tone was also gone. Encouraged by
this success, I ordered two more Nexus fans, so I could replace all the
DustProofs.
REMOVING THE TOP GRILL AND FRONT DOORS
SPCR member theyangster
asked: Why don’t you cut out the fan
grills?
Not knowing why I would want to, I looked around and found this
intriguing entry in the forums (“Case fan grills are evil…”), which convinced me that a grill close to
a fan
creates a lot of noise, and even if it isn’t close it restricts airflow
considerably. Since I want my system to be both finger and paw proof,
as well as filtered, my options in this space were somewhat limited. I
decided to remove the top case fan grill only. Here is a photo. The
copper on the top left and at the top is the VRM heat sinks. The grey
blob at the top right is the temperature sensor of the fan controller
(described below).
Top fan
after the grill is removed; note VRM heat sinks.
The original P180 spoiler
provides safety, and fits without any mods over the soft-mount screws:
P180 top
case fan spoiler in place: no fingers or paws can get to the fan.
The discussion of air flow
restriction also led me to remove the front swing-out doors over the
P180 dust filters. This is easy: just flip each one open, then press on
the
tab near the top hinge. Removing the top fan grill and front doors is a tiny incremental change, but when the goal is very low noise levels and fan speeds, every bit helps.
Maybe some day if I’m feeling both ambitious
and destructive, I’ll cut out most of the metal behind the dust
filters. The Kama Bay would also work better with much larger holes.
USING NOISE MAGIC NMT-3 FAN
CONTROLLERS
The same Edwood
thread in the SPCR forums that got me started on ducting
also talks about the Noise
Magic NMT-2
and NMT-3
fan controllers.
These handy
little devices are designed to be mounted on a corner of an exhaust fan
and control its voltage based on the air temperature. The
NMT-2 and NMT-3 ramp smoothly between 5V and 12V, from
30ºC-50ºC, and
28ºC-42ºC, respectively. I bought one of each to try
out on the top
case fan. Unfortunately the NMT-2 didn’t work (the sense signal never
worked, and after a while every fan I tried would start but then slow
to a halt). The NMT-3 worked fine. After some experiments, I ended up
with it mounted on the front/far corner of the top case fan as shown in
the above photo, where it is sampling a mix of VRM and motherboard warm
air. This results in a speed of 720 RPM at idle, 810 RPM while
folding, and 860 RPM running two copies of CPUBurn. I found this
acceptably quiet, whereas other positions resulted in fan
speeds that were noisier than I liked.
One feature of
the NMT devices is that they spin up the fan at 12V then drop to the
variable voltage. I noticed while experimenting with Nexus fan voltages
and
speeds that these fans do not always start at 5V, although once started
they continue to run. Accordingly, I ordered another NMT-3 to control
the
power supply/hard disk fan in the lower chamber. This lets me run that fan
at 5V which results in an inaudible 545 RPM. This gently wafting breeze
is adequate to keep the disks below 38ºC and the power supply
heat
sink
below 45ºC.
As shipped,
the Nexus fan is designed to be run at 12V only, either from a 3-pin
cable providing sensing, or from a 4-pin Molex without sensing. It
didn’t take me long to replace all that with a standard 3-pin connector
that could be attached to a motherboard plug, an NMT, or an AcoustiFan
voltage adjustor (of which I now had several). When used with an NMT,
the wiring is quite tidy, as shown in this photo of the lower chamber fan.
For the top case fan, I made the fan cable just long enough to reach
the NMT so it wouldn’t dangle into the system.
Lower bay fan with NMT-3
installed and cabling tidied up.
I tried the
CPU and motherboard fans for a while at 12V, and although the cooling
was very good, I found it to be too loud. After a few experiments with
the AcoustiFan adaptors I had lying around, I settled on 720 RPM for
the CPU fan (103 ohms, 7.1V) and 840 RPM for the motherboard fan (78
ohms, 7.9V). With the case closed, the only fan I can hear is the top
case fan, and it is about the same loudness as the buzzing of the
desktop CRT.
THE FINISHED SYSTEM
At last I
could declare victory. A 325W system completely stable, very tidy, and
almost
totally silent. And did I mention fast?
Here’s a photo
of the back panel, showing the video card duct:
Back panel of the finished
system; the top slot is the video card duct.
And one of the front
panel, showing the Kama Bay and dust filters:
Front panel of the finished
system: Kama Bay at the top, P180 dust filters below.
And finally one of the
guts of the system:
Inside the finished system: Four fans, three ducts, two baffles, and lots of copper and foam. The final fan speeds are 545, 720, 720 and 840 RPM at idle, and 545,
720, 840 and 860 RPM under full load. The only way I can tell the
system is making noise is to get right beside it, or to turn it off.
And yes, the guy who got me started on all this was right. It was “tons
of fun”.
* * *
Editor’s Note: Much thanks to Chris Thomson for sharing his experience with us again.
Chris Thomson’s moniker in the SPCR forum is cmthomson, and he may be reached by email at cmthomson at comcast dot net.
* * *
Chris Thomson’s original article: Quiet
DIY OC’ed Pentium D 830 System, Part One
Other Interesting DIY Articles at SPCR:
Doug’s Quiet Wooden Case PC
Quiet A64 X2-3800+ PC for Torrid Thailand
Ducted Zalman 7000CU on A64-3000 w/Countercurrent Flow Cooling
Fanless Heatpipe CPU PC System by FMAH
* * *