Viewing page 4 of 6 pages.
Previous 1 2 3 4 5 6 Next
QUIETLY COOLING THE SYSTEM
|
SIDEBAR: CPU TEMPERATURES
The maximum rating
for
the 830 D is 69°C case temperature, measured at the center of
the case. The thermal diode reports die temperature at a spot centered
on one edge of the die. Two critical parameters are needed to
understand how the die and case temperatures are related: the power
consumption and the junction-to-case thermal resistance.
Intel provides neither of these in the data sheet. Assuming 150W
maximum power and a junction-to-case thermal resistance of 0.12
(typical for this type of
package), the diode-to-case temperature delta should be about
18°C, meaning that the maximum measured temperature should be
87°C.
To sanity check of all this, I measured the temperature of
one of the Ninja heat pipes near the base while the CPU was running
CPUBurn at
a reported 73°C. The outside of the heat pipe was 45°C.
This implies a
temperature delta between the case and the heat pipe of 10°C,
which
seems reasonable for a
high-quality thermal compound and copper heatsink.
As a safety feature, the 830 D implements a second
thermal diode (at an undisclosed location) that controls a clock
throttling mechanism. The activation temperature of this feature is
apparently another secret, although one time when I had screwed up the
CPU cooling, the BIOS reported 103°C.
The highest CPU temperature I saw
during proper testing was a toasty 89°C. This may sound
alarming, but ThrottleWatch never
observed throttling, and I never got anomalous benchmark numbers.
Typical temperatures were in the
mid seventies.
All of these temperatures are way
higher than numbers I see in other people's reviews. I suspect the
thermal diode circuit in my system is out of whack. When I tried
the SPCR
calibration test, the results were so nonlinear that the
formulas didn't work at all.
|
As described in the assembly section, the Ninja CPU cooler had
three 120mm fans
around it: the top and rear case fans, plus a fan blowing up through
the heatsink. The power supply and disks shared a single 120mm fan. My
next task was to find the best speeds for
these fans to provide good cooling at the lowest noise levels.
Cooling the CPU was the most obvious concern, since it uses
most of
the
power in the system. How much power does this CPU draw? Well, that
appears to be a State Secret. The "thermal design power" is 130W, which
implies that the maximum power is higher than that. Using a PowerAngel,
I measured the total system AC consumption at 262W running two copies of CPUBurn. A review
of the 840 EE states that at 3208 MHz and 1.408V it consumes 178.8W. CPU-Z measured my CPU at 1.22V average, and 3456 MHz. Plugging these values into the "V squared x f " formula gives 145W. Directly measuring the 12V AUX wires with
a current clamp meter shows 149W with two copies of CPUBurn.
In the most demanding situation, the CPU, graphics card, memory, and voltage converters consume about 200W. That's a lot of hot air to
try to push quietly through two 4-inch holes. As testing would show, the main challenge would be cooling the motherboard.
Okay, enough theory. Time to talk about fan speeds.
The 120mm AcoustiFans come with a three-speed wiring set that
runs the fan at 12V (black connector, rated at 1500 RPM, 56 CFM and 25 dB),
5V (white connector, rated at 675 RPM, 25 CFM and 10 dB), or an
intermediate voltage connected to 12V through a series resistor (blue
connector). As shipped, the resistor is 56 ohms and the
fan is 92 ohms, so the blue connector provides 7.5V and is rated
at 1050 RPM, 42 CFM and 16 dB. Replacing the inline resistor allows customization of
the fan speed, which varies fairly linearly with the voltage.
I tried a bunch of different configurations and speeds,
including no rear fan, rear fan blowing in instead of out, fast CPU
fan, fast top fan, fast HDD/PS fan, etc, etc. In general, the only
configurations that kept the motherboard from overheating had the rear
fan blowing out, and the top and power supply fans running fastest.
By far the noisiest fan in the system is the hard disk/power supply fan in the lower chamber. This is because the chamber resonates. The typical sound is a
low growl accompanied by a 400-500 Hz tone, depending on the fan speed. Foam doesn't help. The only time this
fan could be considered quiet is when it is run at 5V (675 RPM). At
this speed, it makes a faint hissing sound, and is quieter than the
disks. Unfortunately, when running that slow, a lot of the heat from the power supply eventually transfers
to the upper chamber, causing the motherboard to overheat and become
unreliable. After many experiments trading off noise levels against fan
speed, I settled on a 78 ohm resistor, which runs the fan at 1090 RPM.
The top fan is the second loudest. When I mounted it with the
silicone gasket, I was surprised that the overall sound increased.
This was because the whole center of the fan started to vibrate, creating a
nearly pure 466-Hz tone at 1015 RPM (obtained with a 103 ohms
resistor). I thought I could damp this by wedging some foam between the hub of the fan and the
center of the case opening. This was only partially effective. Quite by
accident, I discovered that touching the center of the case opening
with something metallic (like a screwdriver) got rid of the ringing.
[Those of you who have ever done EMI testing will recognize the
tape-and-tinfoil nature of this process.] So I built a stack of
quarters held together by paper glue to sit on the top of the case and
damp out the ringing. I am not making this up! Here is a recording
where I alternately add and remove the stack of coins, and here is a photo of the setup.
Top fan damped with silicone gasket, foam, and a stack of coins.
With the case closed, the sound from the fan attached to the CPU heatsink is very well damped. I could run it at up to 1100 RPM without adding a
significant amount of overall noise. These higher speeds resulted in CPU temperatures 2-5°C cooler, but had the counterintuitive effect of raising the motherboard temperature. A series of experiments revealed that running the top case fan faster than the CPU heatsink fan worked best
for overall system cooling.
A side note on the power supply internal fan: when the
Phantom 500 was first
reviewed by SPCR, there was some confusion about when this
fan would activate. My experiments showed that it starts to cycle on
and off when the top heatsink of the power supply reaches
51°C, and to spin continuously when the heatsink reaches
52°C. The only way I could get the power supply that hot was by
turning off the center 120mm fan; running that fan at even the 5V
setting keeps the power supply heatsink below 45°C,
so the internal fan never turns on. This is good, because it is rather
loud, and makes a grating ratcheting kind of racket.
At the conclusion of all these experiments, I had set the
lower chamber fan to 1090 RPM (78 ohms), the top case fan to
1010 RPM (103 ohms), the back case fan to 660 RPM (5V), and
the CPU heatsink fan to 690 RPM (5V). Reported CPU temperatures varied
from 61°C idle to 88°C running two copies of CPUBurn.
The system was highly reliable and very quiet. The only
annoyance was a faint 495-Hz tone that had various nodes and antinodes
in the room; by moving the system around a bit, I was able to position
a node where I normally sit at the desk. Aside from that tone, the sound was a low growl, easily ignored.
MEASURING THE SOUND LEVEL
During the quieting process, I became interested in measuring
and recording the various sounds. One nifty tool I found was this
web site, which makes it easy to tell the frequency of a sound you're hearing.
To properly record or measure a system this quiet, you really
should have some professional gear, such as that found in MikeC's
sound lab at SPCR. I wasn't about to spend the several thousand dollars this
would take, but I thought I'd give it a go with some decent amateur
equipment. For $200, I bought a Radio Shack dynamic microphone
and stand, an M-Audio Audio Buddy preamplifier,
cables, and a CEM DT-805 30-dB sound
meter. This collection was almost, but not quite, up to the task at hand.
My $200 "sound lab".
The sound meter shows
readings as low as 23 dB, but these are neither repeatable nor
believable. Taking measurements closer to the fans so that readings
were well above 30 dB was really the best I could do, but to get
readings that high I had to be only a few inches away from one of the
fans. Oh well, another tool for the shed. :-)
AcoustiFan specs the 120mm
fan at 16 and 25 dBA at 1050 and 1500 RPM. The lower chamber fan was running between
these speeds, but also resonated. The overall audible system noise was
less than that of a single fan at 12V/1500RPM, which leads me to guess that
the overall sound level was about 22-24 dBA. Certainly
it couldn't be heard if there was any noise in the house (furnace
running, TV on, wife at home, cat running around, etc). It was much quieter than the
keyboard, or even mouse clicks.
Here is a recording
made from the desktop, about 4 feet from the system. I clicked the
mouse a few times for purposes of comparison. Note that the microphone
is only rated at -70 dB, and the high-frequency hiss you hear is an
artifact of the microphone/preamplifier/computer; the gain was set to
maximum on both the preamplifier and the computer microphone input. The
low growl/hum component is what I actually heard.
| Help support this site, buy from one of our affiliate retailers! |
|
