Chris Thomson returns with a Core 2 Duo upgrade of his quiet PC, greatly overclocked with carefully chosen high performance parts, modified judiciously, and meticulously ducted for maximum airflow and cooling with minimum noise. It’s another magnum opus on the current state of DIY, enthusiast, air-cooled, high performance, silent computing.
Oct 2, 2006 by Chris Thomson (cmthomson at comcast dot net)
Chris
has been a
computer system architect at Myrias Research,
Amdahl, 3Com, Nokia and others for the last 20+ years. An ex-pat
Canadian, he now resides in sunny Pleasanton, CA. Chris used
to think that people who build their own PCs and then overclock them
are nuts. Over the last year or so, he’s become convinced of it.
Chris caught the silent PC bug while building a hot Pentium D 830 dual-core system a
year ago, and has become an active SPCR member. This article describes a complete upgrade to an overclocked Core 2 Duo system that is quiet enough to make his laptop PC seem outrageously loud.
* * *
TIME FOR AN
UPGRADE
After my excruciating effort to quietly cool the 150W space
heater
marketed under the moniker Pentium
D 830 (documented on SPCR in two
parts, Quiet
DIY OC’ed Pentium D 830 System and Quiet
DIY OC’ed Pentium D 830 System, Part Two),
California summer arrived, and the temperature-controlled fans in my
system kept going faster and faster and got noisier and noisier. I
started to get antsy about upgrading my system to the next generation
CPU, both for lower wattage and for better performance.
In the last year, three major events have impacted
quiet high-performance PCs: the introduction of the Core
2 Duo “Conroe”
CPUs from Intel, the relentless race by video card GPU vendors for
higher performance regardless of power consumption, and the catch-up
race of heat sink vendors to cool those GPUs. Two good, one not so good.
Of these, the Conroe is probably the most significant. Not
only
is the power consumption far lower (as much as a 2/3 reduction), but
the performance has been boosted substantially, particularly for audio
and video transcoding, the most common compute-bound tasks I do. This
is because the cache is very large (4 MB on the E6600) and the
execution units have twice as many SSE stations as previous CPU cores.
The "core" of my upgrade.
The performance increase in GPUs is also impressive: the newer
7900GT
outperforms my old X800XL by almost 2:1, and it is not even a top
performer anymore. It is, however, a very good compromise of
performance, price, and power consumption.
Time has marched on for memories as well. My old 5400C4 memory
was
looking pretty sad next to newer 6400C3 sticks. Even my old DVD writer
was looking tired. And of course it was time to replace that CRT with
an LCD.
In revamping my system, I set the bar high for myself: a
system at least twice as fast as the
old one, and so quiet it couldn’t be heard at 6am on Sunday. By
starting with a large well-damped case, using
gigantic heat sinks and very slow fans, minimizing the system airflow
impedance, ducting hot air directly
out of the box, and overclocking fast components to almost
overload the heat sinks, I was completely successful.
MY NEW SYSTEM CONFIGURATION
As the old carpenter said: “I’ve had that same hammer for 25
years, although I’ve gone through four handles and two heads.” I still have the same system, it just has a new CPU,
motherboard,
memory, video card, DVD burner and monitor. Here is the complete
configuration:
- Antec P180 case (the original silver/black one)
- Antec Phantom 500 semi-fanless power supply
- Intel Core 2 Duo E6600 CPU
- Asus P5W DH Deluxe motherboard
- 2 GB G.Skill F2-6400PHU2-2GBHZ memory
- eVGA 7900 GT KO SC (N567) video card
- ATI TV Wonder Elite TV card
- Samsung 244T 24″ LCD monitor
- two Samsung SpinPoint SP2004C 200-GB SATA disks
- Sony DRU820A DVD burner
- generic floppy drive
- Scythe Ninja CPU heat sink
- Thernalright HR-05 north bridge heat sink
- Aerocase Condor GPU heat sink
- Swiftech MC14 copper RAMsinks
- Scythe Kama Bay filtered air inlet
- Nexus Real Silent 120mm orange fans
- Noise Magic NMT-3 temperature-sensitive fan controllers
- AcoustiPack foam
- home made styrene ducts and baffles
- various peripherals (UPS, printer, zip drive, speakers,
keyboard, mouse, etc)
| In the rest of this article, I’m going to document the system as though it had been built from scratch. In actuality, most parts were carried forward unmodified from my old 830D configuration. Most of the concepts are also documented in my earlier 830D articles, along with credits to SPCR members who suggested them. |
COOLING THE CPU, NB
& VRM: THE FANTASY
The E6600 is rated at 65W TDP (thermal design point), and
actually
consumes closer to 55W. I knew that cooling it would be
straightforward, but experience
with my old 830D/P5LD2 configuration left me concerned about
cooling the motherboard components, especially the Vcore VRM (voltage
regulator module). And of course the 95W video card would need some
attention.
Two of the new Conroe-ready motherboards caught my eye: the
Gigabit
GA-965P-DQ6 and the Asus
P5W DH Deluxe. Both feature heat pipes from
the north bridge to radiators near the CPU, as well as radiators for
the VRM MOSFETs, also near the CPU. In addition to good VRM cooling,
these boards also use highly efficient multiphase VRMs (12 and 8 phase,
respectively).
This made me think that I could cool the CPU with a
Thermalright
XP-120 heat sink with the fan blowing up from the
motherboard, which
would force cool air through those north bridge and VRM radiators. I
would then duct the warm air straight out one of the case holes. This
would eliminate the need for a case fan, since the motherboard and CPU
would be sharing a fan. The video and TV cards would be cooled
separately.
I initially tried the DQ6 motherboard, but it was not a happy
experience. The 965P north bridge actually has fewer features than the
945P on my old motherboard, and would not allow me to directly control
the DRAM parameters. Also the BIOS is set up for overvolting and
overclocking only; it is not possible to set voltages directly, or
below the stock values called for by the VID and SPD. I ended up
selling the board and switching to the P5W DH, which uses a 975X north
bridge and has the features I wanted.
With the P5W DH, the XP-120 can be mounted in the “standard”
orientation, with the heat pipe bends toward the DIMMs, or rotated with
them facing the top of the case. The latter is better, as it provides a
larger space between the heat sink and the case outlet, allowing for a
larger duct with smoother bends. It is also easier to mount in this
direction because there is more room around the clips. Here is what the
XP-120 looked like installed in the P180.
XP-120 CPU heat sink mounted on
the P5W DH inside the P180 case.
Note how the heat sink lines up quite well with the hole in
the
case. I never actually built the duct, but it would have been quite
simple. Note also that I ran the CPU power cable under the motherboard.
COOLING THE CPU, NB & VRM: THE REALITY
It was a pretty good plan, and it would have worked if I
hadn’t given in to the dark side and started overclocking heavily.
This required boosting the CPU core voltage, which in turn boosted the
power consumption radically, into the vicinity of 100W. As
with my old 830D, this much CPU wattage meant unacceptably loud fan
speeds to obtain reasonable temperatures with an XP-120.
Since I already had in hand a Scythe
Ninja heat sink and matching duct from my previous
build, I switched to that heat sink, which I knew for certain could
cool the CPU quietly. The Ninja is huge and efficient. Here is a photo
from its SPCR review.
The Scythe Ninja tower heat sink,
with a 120mm fan attached.
While running the CPU at high voltage, I noticed that the VRM
was
getting pretty warm, though not nearly as hot as on my old
motherboard.
Still, it was hot enough that the heat pipe to the north bridge
was actually working backwards: it was transmitting heat toward
the north bridge instead of away
from it.
My stockpile of earlier experiments contained a Thermalright
HR-05
tower north bridge heat sink, so I decided to try it instead of
the stock heatsink/heatpipe/radiator thingie. This was not particularly
easy because although the HR-05 comes with a clip designed for
Intel-style hoops, it is designed for hoops on
the northeast and southwest sides of the north bridge. For some
bizarre reason, the P5W DH has provision for all four hoops but only
has two of them populated: the northwest and southeast ones. Mounting
the HR-05 required severe bending of the clip to reverse the angles,
since I didn’t want to take a soldering iron to this insanely expensive
and hard-to-get motherboard.
When mounted as designed, the HR-05 can be rotated to various
angles
as needed. In my reversed orientation, the attachment clip bumped into
the bottom fins, and prevented this rotation. To address this, I
cut notches in the bottom two fins, as shown in this (very
blurry) close-up photo.
Notches in the bottom fins of the
HR-05 allow it to be rotated.
The HR-05, though billed as a passive heat sink, does need
some airflow to cool an overclocked 975X. Also, since
the Ninja is a tower heat sink, there is no downward airflow onto
the VRM from its fan. Fortunately, the fan on the Ninja can be
positioned so that it pushes some air between the Ninja and the
motherboard, and also between the Ninja and the HR-05. To capture as
much of this air as possible, I angled the HR-05 with respect to the
Ninja. This close-up shows the fan and the two heat sinks.
The CPU fan spills some air onto
the angled HR-05 heat sink.
Because I planned to duct the CPU heat directly out the back, there needed to be a top case fan to cool the rest of the motherboard. To minimize
noise, I wanted to soft-mount this fan, but the P180 case is
not designed for soft-mounting a top case fan. The top fan mount has
two screw holes
at the back, but at the front instead of screw holes it has
two bent metal tabs to hold the
edge of the fan. I got around this by breaking off those tabs and
drilling two new holes. This is easy, since this part of the case is
plastic. Just use a fan as a drill guide. I soft-mounted the top fan
with an AcoustiFan
gasket. After cutting out the grill to reduce noise and
mounting the fan, the P180 spoiler still fits in place, and looks like
this:
The P180 top spoiler fits over
the soft-mount screws to keep fingers and paws out.
Here is an inside view
of the CPU and north bridge heat sinks installed, together with the CPU
and top case fans.
Ninja CPU and HR-05 NB heat sinks
installed, with CPU and top case fans.
When the CPU fan is positioned flush with the heat sink fin
closest
to the viewer in the photo above, it intrudes into the first DRAM
slot. Since I have only two DIMMs, this is fine. I use the black
connectors, which are farther from the CPU.
The sides of the Ninja line up very well with the case
openings,
making ducting of its exhaust air very simple. This was mandatory in my
old configuration, and still worthwhile in my new one. The duct I use
has four flat panels, two long and two short. The long panels seal the
top and bottom sides of the Ninja and force the air out the back case
opening, which is sealed with some foam strips (these also bear some of
the weight of the Ninja, its fan, and the duct). You can see the foam
strips on the left in the photo above. Here is what the finished Ninja
duct looks like, as viewed from the back of the case.
Ninja duct (rear view), with its
foam lining.
The back portion of the duct is lined with thin (4mm) foam.
This is
not neccessary with a slow fan, but when I first built this duct I was
using pretty fast fan settings and wanted to absorb some fan noise. The
duct slips over the Ninja and nestles in the foam strips at the back of
the case, like this:
Ninja duct installed in the
system; air still gets to the HR-05.
Like all of my ducts, this one is made from styrene, a stiff
plastic that is very easy to cut and glue. You can find a tutorial here
(posted by a guy in my home town!).
The base of the Ninja (below its fins) stands two inches above
the
motherboard. My duct covers only the fins, so it is possible to force
air between the duct and the motherboard, with the effect of cooling
the VRM. The approach I use to do this is to pull air across the board
with the top case fan. To direct this air flow properly, I built a
two-panel baffle that mates with the Ninja duct and
the case walls to seal off all other paths to the top case fan, as
shown in this photo.
Baffle to force top case fan air
flow across VRM.
Here is what it looks like installed, with its edges tucked
into some foam. As you can see, the only way
for air to get to the top case fan is from the area around the base of
the Ninja. This cools the VRM.
VRM baffle installed, together
with Ninja duct.
Because the Ninja and HR-05 are remarkably efficient, and the
exhaust air is ducted directly out of the box, only the gentlest of
breezes is needed to cool the CPU, north bridge, VRM and DRAM.
So with two fans, two monster heat sinks, and a couple of
ducts, the top third of the system is take care of.
MOTHERBOARD IDIOSYNCRASIES
The Asus P5W DH Deluxe motherboard is a top-of-the-line
product, and is priced accordingly. It has been the favorite of
Conroe overclockers, and at the time of this writing (mid-September
2006)
has been on constant backorder for over a month due to this popularity.
It does, however, have some oddities, many of them related to
cooling and overclocking.
The most glaring problem is the shiny copper-colored covers on
the
north and south bridge heat sinks (the ones that say “Digital Home” and
“ASUS”). These must
be removed; they completely obstruct any airflow through what are
otherwise pretty good heat sinks. They are very easy to remove.
Second, the stock south bridge TIM is low quality and really
should
be replaced. This is fairly straightforward. With the board out of the
case, use needle-nose plyers to squeeze the push pins on the back of
the board, which allows them to slide through the holes in the
motherboard.
Then use a large flat screwdriver levered on the blue IDE connector to
pry up the south bridge heat sink. It will let go with a small “pop”.
Pry off the south bridge heat
sink to replace its TIM (release the push pins first).
Here you see why this TIM should
be replaced.
Next, replace the TIM. As always when replacing TIM, clean the
chip and heat sink
thoroughly then apply the smallest amount of new TIM that will cover
the chip. The push pins snap through the board and hold the heat sink
firmly in place. When done, the heat sink looks much more competent:
South bridge heat sink with cover
removed.
If you plan to keep the stock north bridge heat sink, you
should
replace its TIM as well. The heat sink and related VRM radiator are
held onto the board with four push pins. Take care while cleaning the
heat sink not to damage the foam spacer that fits around the north
bridge chip.
There is a fully illustrated how-to for the above steps here.
The temperature readings on the P5W DH are very unusual. The
temperature reported by the supplied PC Probe 2 software, and by SpeedFan,
as “motherboard” is actually an on-chip sensor inside the south bridge.
The temperature reported as “CPU” is a sensor on the board near the CPU
socket, not the on-chip thermal diode or the DST (digital thermal
sensor,
a new feature in the Core 2). As a result, the reported motherboard
temperature is unusually high, and the reported CPU temperature is
exceptionally low. Also, there is a bug in Probe 2 (or perhaps the
BIOS) that causes it to occasionally report a CPU temperature of 256C.
There are two reasons why the on-board CPU temperature is well
below the actual CPU chip temperature. One is that the motherboard has
a large heavy copper layer around the CPU socket (called Stack Cool)
that draws heat away from the socket very well. The other is that the
conduction from the CPU die to the motherboard is quite poor, as shown
in this illustration from the data sheet:
Why a sensor in the socket can’t
accurately report CPU temperature.
The IHS (integrated heat spreader) in the above picture is
nickel-plated copper, and is designed to transmit heat to the CPU heat
sink, which is mounted above it with an external layer of TIM in
between.
The on-chip CPU DST can be accessed with Core
Temp, RMClock,
or Everest
Ultimate Edition,
and is highly accurate. However, it is reported as a negative number
relative to the (secret) temperature at which automatic throttling
starts. Core Temp and Everest use 85C for this temperature, RMClock
uses 97C. In some sense, the absolute temperature is not relevant; any
good CPU cooling system will prevent throttling. All of these
temperatures are well below the 105C junction temperature that
semiconductors are designed to withstand.
The remaining issues with this board are with its BIOS. First,
very
early versions of the BIOS would not POST with Conroe CPUs. This was fixed
in release 0801 and all later releases.
Second, many Asus boards overclock the north bridge, a
feature called Hyper Path 3 in the BIOS. On the P5W DH, this is
done by tying the CPU to the 1066 strap and the north bridge to the 800
strap. This reduces the latency through the north bridge, but causes it
to run very hot, and also limits how fast the FSB can be run. To
successfully overclock this board, Hyper Path 3 needs to be disabled.
There is a detailed explanation of 975X overclocking here.
Third, it is not possible to change the CPU Vcore voltage
unless
Enhanced C1 and SpeedStep are disabled in the BIOS. This is made
explicit in later BIOS versions, but many people had trouble early on.
COOLING THE HARD DISKS AND
POWER SUPPLY
The P180 case design separates the power supply and
motherboard air
flow using two chambers. The hard disks can be mounted in either
chamber. The lower chamber has a fan in the middle that pulls air
across the hard disk cage and pushes it across the power supply. I put
my disks in the lower chamber.
I chose the Antec
Phantom 500
“semi-fanless” power supply for my system, knowing that its fan would
never come on as long as the case fan was running. The P500 is a
revision of the Phantom 350, which is completely passive.
This makes the P500 larger than a standard power supply, which requires
that the center fan be at most 25mm thick. The stock fan is 38mm thick,
and very noisy, so it was destined to be replaced by a Nexus.
To slow this fan down, I installed a temperature-sensitive NMT-3
fan controller from Noise
Magic.
This controller has an output ranging from 5V to 12V as the temperature
ranges from 28C to 42C when connected to a 12V source, and passes the
tachometer signal through to the motherboard. Regardless of
temperature, it outputs full voltage at startup, to get the fan
spinning. This is needed because Nexus fans do not start reliably below
6V. (I also put an NMT-3 on the top case fan.)
Here is what the lower chamber fan and its NMT-3 look like
mounted on the mid-chamber baffle, ready for installation.
Lower chamber fan, NMT-3
controller, and baffle.
The Phantom 500 derives most of its cooling from air flow
across its
top heat sink. To maximize the effectiveness of this cooling, that area
needs to be open, as well as the back part of the case. Consequently,
all spare power supply cables should be tucked under the supply.
Tuck unused power supply cables
in the bottom to keep the top open for cooling.
The power supply fan activates when the top heat sink reaches
51C.
However, the center fan keeps it well below this, even when spinning as
slowly as 400 RPM. The result is silence.
The case has two sliding panels that allow cables to go
between the
lower and upper chambers. The main one for the power cables has a
rounded notch that seals air flow quite well. The second one for the
disk cables does not. This is simply rectified with a notch so it can
completely isolate
the two chambers. Here’s what the slider looks like after being notched:
A notch in the smaller slider
lets the SATA data and power cables through.
So with just one fan and some cable management, the bottom
third of the system is taken care of.
COOLING THE VIDEO AND TV CARDS
At 95W, the overclocked 7900GT video card is almost as
power-hungry
as the overclocked CPU. The eVGA stock cooler works remarkably well for
a single-slot
design, but has two serious issues: it dumps its heat back into the
case
(and recirculates some of it), and it is unbearably loud when
running 3D loads.
The TV card consumes “only” 25W, but as supplied has no heat
sinks at all, and runs extremely hot in a case with low air flow.
My overall plan was to cool these two cards and the
south bridge separately from the CPU, north bridge and VRM by dividing
the upper chamber of the P180 into two subchambers,
resembling the way the power supply and hard disks are segregated from
the rest of the system. With appropriate baffling and/or ducting,
a fan could supply fresh air to this new chamber, and the hot air could
be exhausted directly out the back of the case.
During the last year, some new video card coolers have come
out that are
as manly as the Ninja. In particular, Aerocase launched its Raven and Condor heat sinks,
gigantic affairs targeted at cooling the toastiest GPUs passively. The
Condor can be ordered in a “reverse wing” orientation, like the sample SPCR reviewed, so that its
radiator sits below the video card far enough away from the motherboard
to allow standard PCI cards to be installed under its wing. This fit my
plan perfectly.
My first task in building the video chamber was to add some
radiators to the TV card. I had
a bunch of Swiftech
MC14 copper RAMsinks
available, and started out by putting them on the very hot parts of the
card: the ATI chip, the VRM, and the tuner casing. This didn’t really
do the job, so I proceeded to stick them on every available flat
surface near a hot spot spot on each side of the card.
RAM sinks on the component side of
the TV card…
…and on the solder side.
Next, I installed this porcupine in the system.
TV card and its heat sinks in the
system.
It was now time to tackle the video card and the Condor.
Aerocase is a very small company that currently builds each
heat sink to order. This became evident to me when I phoned the company
to confirm my order (they didn’t have an automated e-mail system)
and the president took my call. She was very interested in how I
intended
to use the reverse-wing Condor, and we had a pleasant chat while I
explained my tentative plan.
When the package arrived a couple of weeks later, it contained
a
slip of paper informing me that my heat sink had been randomly selected
for post-assembly testing, and might appear used. It was true: there
were several scratches on the heat pipes and a couple of fins were
bent. The heat block also had about five times as much Arctic Silver 5
on it as needed. After cleaning this off and putting a fresh dab onto
the GPU, I was ready to install the heat sink on the 7900GT card.
First I had to bend the heat pipes back to 90 degrees. They
had been
bent, apparently during packing or shipping, to about 110 degrees. A
little gentle pressure did the trick. Next, I had to thread the
supplied bolts through the video card into the embedded nuts in the
Condor heat block. The instructions were helpful, if a bit down-home:
put the heat block on a coffee cup, insert the toothpicks, add the
rubber spacers, then lower the video card onto the heat block using the
toothpicks as guides. This worked okay, but one of the four nuts did
not line up with the corresponding hole in the video card; it was off
by a fraction of a millimeter. Using a bit of (careful!) force, I was
able to thread all four bolts.
This revealed the next issue: the supplied red rubber washers
were
much too thin to cushion the heat block on top of the GPU. Here’s a
picture showing this:
The red washers are too thin for
this video card.
I had to very carefully tighten the bolts in a diagonal
pattern to
squeeze the heat block onto the GPU without breaking anything. (What
looks like a warp in the above picture is a lens artifact; the board is
flat.) This cooler is defnitely not for the uninitiated.
Here is what the final assembly looks like.
Video card with Condor mounted.
The uneven spacing of the wing parts is evidence of hand
assembly.
Perhaps the perfect alignment and spacing of Thermalright and Scythe
heat sinks has
spoiled me. Of course none of this has any effect on the actual
performance of the heat sink.
(Note there are no RAMsinks on the board. I used to be a big
believer in these, which is why I had so many available to put on the
TV card. However, as part of the Condor discussion on SPCR (page
4), it was pointed out that GDDR3 chips don’t get hot. It’s
true: the
DRAMs on my card don’t get much hotter than the card itself. That
discussion also pointed to this article,
where a gonzo overclocker used a phase change system to supercool his
GDDR3 and got no speedup.)
The 7900GT with the Condor attached fit nicely into the P180, and wrapped around the TV card as expected.
Condor in the system, wrapped
around the TV card.
By happy accident, the Condor radiator lines up nearly
perfectly
with the rectangular vent in the back of the case originally
intended for the ill-fated GPU duct/box/thing that came with the early
P180s. The warm air exhaust for the video/TV chamber is
comprised of the rectangular vent along with the two holes created by removing the PCI slot covers from above and below the TV card. Note that my case has the old-style solid PCI covers, which I think do a better job of directing air flow than the new-style perforated blanks. Here is a view of the exhaust holes.
Hot air from the Condor and the
TV card goes straight out the back of the case.
The next job was to install a fan to blow fresh air into the
video
chamber. My first attempt was with an Aerocool Streamliner 14cm fan.
This fan fit the opening very well, and I thought its extra large size
would make it quiet and efficient. Unfortunately it was neither. It
moved less air than a Nexus, and was considerably louder.
So even though it didn’t fit as well, I installed a Nexus fan
near
the upper disk cage. The cage is still present, but its caddies have
been removed, as seen here:
Video fan nestled among
connectors and cables.
Although it looks a bit cluttered in the above photo, the air
flow
is almost entirely unobstructed. The fan is sitting down on the blue
IDE connector, wedged between the SATA connector and the drive cage.
The various power cables hold it mostly in place, and the single cable
tie to a taut power cable secures it firmly.
The top side of the video chamber is formed by a single large
flat panel that runs the full length and depth of the chamber.
This large panel forms the top of
the video air flow chamber.
I added the quarter to give the camera something to focus on,
but
also to give an idea of just how big this panel is. The cut notches at
the bottom are to accommodate various chips, capacitors and connectors
on the motherboard. The tab near the center of the bottom edge is a
vertical spacer that rests against the video card motherboard
connector to hold the panel above the card.
Installed in the system, the panel looks like this:
The video chamber, nearly done.
The spacer tab can be seen in the above photo holding the far
edge
up. To the right of that, the panel is threaded between the DIMM
handles and the top of the fan, which hold that end snugly. The left
edge butts against some foam at the top of the exhaust opening. The
close edge is sealed off and held in place by some foam attached to the
case side panel.
The final task was to seal the area around the fan. Two layers of thick foam, one full-length and the other shorter, took care of that:
Foam seal added to the video fan
completes the chamber.
So with one fan, one baffle, a monster heat sink, and some
copper and foam, the middle section of the system is taken care of.
ALL THREE CHAMBERS, ALMOST READY FOR ACTION
Here’s an overall photo showing the three chambers and their
various cooling aparatus.
 The complete system: Four fans, one duct, two baffles, four monster heat sinks on the CPU, NB, GPU and power supply, and some copper and foam are all that’s needed for efficient cooling! |
MAXIMIZING INTAKE
AIR FLOW
The stock P180 case has only one air inlet for the
motherboard chamber, and another for the power supply chamber. To
efficiently feed the three fans in the motherboard
chamber, I needed to expand the inlet cross section.
I had three 5.25″ bays available since I use only a single
DVD/CD drive. This is ideal for installing a Scythe
Kama Bay. The Kama Bay (with its fan removed) more
than doubles the inlet area, and provides filtered air.
Installing the Kama Bay without duct tape requires fitting the
steel P180 drive
clips to its shell, which requires drilling some holes. These clips are
made from a very hard steel; to prevent wandering,
I needed to clamp the clips between two layers of waste material before
drilling.
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.
The last step was 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. 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. Of
course I run the system with the main front door open to minimize air
flow impedance. Because the system sits behind my desk, this doesn’t
cause any issues.
Here is the finished front
panel:
Finished front panel: Kama Bay at
the top, P180 dust filters below.
OVERCLOCKING THE SYSTEM
The Conroe family is very overclock-friendly, as are the new
motherboards and memories. Reports of overclocks of 50% or more are not
uncommon on the enthusiast forums.
The latest generation of CPUs require new test and stress
software. Older programs such as CPU
Burnin, CPUburn,
and Prime95
do not load the CPU fully. The new gold standard for CPU, north bridge
and memory stability is ORTHOS,
which runs two copies of Prime95 with varying parameters, while the
best CPU load program for Core 2 or Core CPUs is TAT (Thermal Analysis
Tool), which is a proprietary Intel program that can be downloaded
through links in this
thread. The best video card stress tools continue to be rthdribl
and ATI
Tool.
ATI Tool works with both ATI and nVidia GPUs, and consumes slightly
more power in tumbling block mode than in artifact check mode.
Running TAT at 100% on both CPUs, the core
temperature of the E6600 at
stock frequency and voltage (2.400 GHz and 1.35V) goes up to only 68C
with the Ninja fan running at just 600 RPM. If you’re not overclocking
this CPU, you really don’t need the kind of cooling described above.
However,
since my plan all along had been to overclock to the maximum frequency
supported by quiet cooling, this temperature serves only as a
comparison point.
I started my overclocking with the memory. The G.Skill HZ
memory is
hand-picked to support high clock rates and low latencies at elevated
voltages. The highest reliable frequency I found for this motherboard
and
memory was 740 MHz 3-3-3-12 at 2.3V, or 812
MHz 4-4-4-12
at 2.4V. As usual, the memory is not the limiting factor in my
system clocking.
Initially, I had a week 24 CPU (the first week of production),
which
required a relatively high Vcore to overclock. My initial target for
fan speeds was 700 RPM, which is very quiet but still audible. I was
able to overclock to 3.244 GHz, but only at rather low ambient
temperatures.
Later I got a week 28
CPU, which overclocks well at noticeably lower voltages. With this
newer CPU, I was able to run the system with FSB
and DRAM frequencies of 361 and 722 MHz, and DRAM, CPU and north bridge
voltages of 2.30, 1.4625 and 1.55. The DRAM parameters were set to
3-3-3-12-5, and C1 Enhanced, SpeedStep and Hyper Path 3 were disabled.
This ran the CPU at 3.252 GHz, and was completely stable with the fans
running at only 600 RPM.
Getting the last few drops of performance required adjusting
the tRD memory parameter. This parameter is not user-settable in the
BIOS, which silently sets it to 5 when the other parameters are 1:1 and
3-3-3-12-5. Changing it to 6 with memset
3.0 made my system stable with FSB and DRAM clocks of 370 and
740. This runs the CPU at 3.333 GHz. To achieve complete stability, I
had to overvolt the DRAM, CPU and north bridge to 2.30, 1.50 and 1.55
volts. Higher overclocks are possible, but only with faster fan
settings. I was so pleased with the inaudibility of 600 RPM that I
decided to stick with that.
Recent versions of the BIOS disable CPU auto throttling
(called thermal control in the user interface) by default. If you want
your CPU to
last very long, it is imperative that you enable this feature in the
BIOS, or that you always run RMClock
to provide throttling. I let the BIOS control it.
eVGA sells many variants of the 7900GT video card. The N567 7900
KO SC
(KnockOut SuperClocked) is overclocked at the factory to run the GPU
and GRAM at 580 and 790 MHz. As with any nVidia card, to display or
modify this clocking, you must install the CoolBits registry hack,
available here.
To achieve this clock rate, the GPU is overvolted from 1.2V to 1.4V
when running 3D applications, which causes it to consume much more
power than a standard 7900GT. After several experiments I was able to
increase this overclock only slightly; in the end, I reverted to the
factory settings.
I ran a bunch of benchmarks to compare my final overclock with the
stock CPU and memory settings, and also, as much as possible, to my old
830D system.
| Benchmark | Stock E6600 | OC E6600 | OC 830D |
|---|---|---|---|
| FSB, DRAM, CPU clocks | 267, 333, 2.404 | 370, 740, 3.333 | 235, 627, 3.532 |
| Video GPU, DRAM clocks | 580, 790 (1580) | 580, 790 (1580) | 427, 555 (1110) |
| DRAM latencies | 5-5-5-12-9 | 3-3-3-12-5 | 4-4-4-12-4 |
| Sandra iSSE, fpSSE | 132345, 71415 | 184027, 99278 | 35260, 40143 est |
| Sandra RAM bandwidth: int, fp | 5204, 5215 | 7214, 7226 | 5820 est |
| Sandra memory latency | 90 ns | 61 ns | – |
| 3DMark05, 3DMark06 | 10134, 5609 | 10406, 5911 | 5379, – |
| PCMark04 CPU, memory, graphics, disk | – | -, -, 11273, 12307, 4830 | 6999, 5569, 6795, 4372 |
| PCMark05 CPU, memory, graphics, disk | 6082, 5165, 8283, 4246 | 8472, 7249, 9104, 4256 | – |
| PCMark04, PCMark05 system | -, 6580 | -, 8180 | 7114, – |
This system is very quick. Note the outstanding SSE scores; this
system encodes video over three times faster than my old one. Its video
FPS rate is nearly twice as fast.
FINALLY, SETTING FAN SPEEDS
As hinted many times above, I run my fans very slowly.
The Ninja is a truly exceptional heat sink. Even with the CPU
consuming 100W at this extreme overclock, and the fan spinning at an
inaudible 600 RPM, the cores stay well below throttling under my
heaviest realistic workload (video recoding plus protein folding). This
workload corresponds to 75% loading in TAT. With faster fan settings it
is possible to avoid throttling even under the synthetic workloads, but
I stuck with 600 RPM since it was all I needed. Because various programs report throttling temperatures varying from
83C to 97C, I can’t say with any precision what my actual CPU
temperature is.
The Condor cools the GPU very well, needing only a slight
draft to shed its heat. I ran a series of fan speed tests loading the GPU with
the ATI Tool artifact scanner. This provides a higher load than any
game I’m aware of, though slightly less load than rthdribl. I used
SpeedFan to adjust the voltage to the Nexus fan installed in the video
chamber, and measured both the GPU and south bridge steady-state
temperatures. The ambient temperature was 22C.
| Nexus 120 Fan in Video Card Chamber | ||||||||
|---|---|---|---|---|---|---|---|---|
| SpeedFan % | 44 | 50 | 55 | 60 | 65 | 70 | 75 | 80 |
| Fan RPM | 598 | 670 | 721 | 774 | 830 | 870 | 917 | 970 |
| GPU – °C | 105 | 104 | 98 | 94 | 94 | 92 | 89 | 88 |
| SB – °C | 43 | 43 | 41 | 40 | 39 | 39 | 38 | 37 |
The fan becomes barely audible at about 650 RPM, although it
is very quiet by most people’s standards even at 1000 RPM. In this
setup, there is no reason to run the fan faster than about 750 RPM. The
default throttling temperature for this GPU is 130C, so none of the
temperatures here are particularly excessive.
Nonetheless, in order to keep the GPU below 100C, I ultimately
decided to let the fan run faster under load. I set SpeedFan to use a
minimum of 44%, a maximum of 55%, and a desired south bridge
temperature of 42C. The only time the fan speeds up is during
benchmarks.
The P5W DH has five fan headers: CPU, CHA1, CHA2, PWR1 and
PWR2. As the naming implies, these form three groups from SpeedFan’s
perspective: CPU, CHA and PWR. I used one of the PWR headers to control
the top case fan, and the two CHA headers to control the video and
power supply/hard disk fans. The top case fan and the power supply fan
have NMT-3 controllers; these work well in combination with SpeedFan
and slow these fans even more when temperatures are low.
I set the CPU fan to 43%, the top case fan to 44%, and the
video and PS/HD fan to 44-55% depending on the MB temperature. When the
ambient is 22C and I’m folding, the fans run at 580, 590, 590 and 480
RPM. They speed up a bit when the room is warmer, typically 20 RPM. The
hard disks stay below 40C, and the power supply fan never spins up.
How quiet is this system? Well, if the DVD isn’t spinning and
the hard disks aren’t seeking, the only perceptible sound is the faint
hum of the LCD power supply echoing off the wall six feet away.
Obviously the house and neighborhood must be totally quiet to hear
this. By comparison, my company laptop is outrageously loud.
* * *
Articles of Related Interest
Quiet DIY OC’ed Pentium D 830 System, Part Two
Quiet OC’ed Pentium D 830 System
Quiet PC for Torrid Thailand
Doug’s Quiet Wood Case PC
* * *
