Viewing page 6 of 6 pages.
Previous 1 2 3 4 5 6
TIME TO RESORT TO DUCTING
I continued to be bothered by the lack of cooling for the MCH. The
heatsink was too hot to hold a finger on, which is not a good thing. I
decided to search SPCR for ducting ideas, and came across Edwood's article about his HTPC. In addition to some fascinating pictures of custom ducts, it also pointed to this tutorial on styrene cutting and cementing.
Since the P5LD2 MCH heatsink has a pair of slots cut horizontally, I figured
I could make a duct that would slide into those slots and divert airflow
from the motherboard fan directly onto the heatsink. The idea was to
have the duct friction-fit into the heatsink and bump against the back DIMMs,
just downwind from the motherboard fan. I actually got it right on the first
try, although it's a bit ragged. Here's what the duct looks like from the bottom (the heatsink
side):
The MCH heatsink duct, outside the system (bottom view).
To make room for the duct, I had to shove the CPU fan upward, which
was a good thing since that increased the air flow across the Vcore
VRM. Here's what the MCH duct looked like in the system, before the
motherboard fan was mounted:
The MCH heatsink duct, inside the system (front view).
This had excellent results. Whereas previously the north bridge heatsink was very hot to the touch, now it was barely warm. The reported
motherboard temperatures also dropped a degree or so.
Encouraged by this success, I started to think about how to vent the
CPU heat directly from the case, separating it from the motherboard
heat. I figured if I could encapsulate most of the Ninja and direct the
hot air out the back, the motherboard cooling would improve. First I
removed the back case fan, which I was no longer running anyway. Next I
built a square duct with flaps to cover the top and bottom of the Ninja
and route the air flow to the back vent. I lined the back portion with
foam, thinking that ought to reduce the CPU fan noise. Here's what it
looked like before being put in the system:
Ninja duct and its foam lining outside the system (back view).
After installation, I put some foam strips on the case wall
by the back of the Ninja duct to keep it in place and seal off the air
flow. Here's what the two ducts and fans look like inside the system:
Both ducts and fans installed in the system.
Although in the picture above it looks like the Ninja duct goes all
the way down to the motherboard, it is actually an inch above it,
allowing airflow all around the base of the CPU tower. Also, the
downwash from the MCH duct flows across the motherboard surface.
While designing the Ninja duct, I was hoping I could augment
the airflow through the VRM heatsink with some kind of baffle on the
bottom, but unfortunately the VRM heatsink is so
close to the CPU that there is no room for any kind of deflector. Pity.
With these two ducts in place, I tried running the system with all
the fans at 5V. Unfortunately this didn't adequately cool the
motherboard at the new clock settings described below; I needed to
return the motherboard fan to 995 RPM (103 ohms).
During these tests I noticed that the entire top of the case was
vibrating and contributing a lot of noise. This seemed to be caused by
the "semi-hard" mounting of the top case fan. I decided to soft-mount
it properly with AcoustiFan gaskets and screws. This required drilling
two holes for the front screws, and bending the inside case tabs
completely out of the way. The fan hub vibrated almost as much as
initially, so I added the foam-and-coin damping described above with
good results. Here is what the soft-mounted fan looks like. The P180
slip-on cover fits over the screws with minimal distortion.
Top case fan soft-mounted with AcoustiFan gasket and screws, damped with foam and coins.
The last step before photographing the finished system was to tidy
up the cables. The AcoustiFan three-speed cable assemblies are quite
unsightly, so I tucked them into an unused disk bay, along with the
PATA cable and power splitters and connectors. I routed the 12V AUX
cable around the top of the Ninja duct. This is what the finished system looks like:
The finished system, with cables tucked mostly out of sight.
This setup is almost silent. When I try to record its sound, all I
get is microphone/amplifier hiss. With the system positioned back
behind the desk, I can hear it only if the house and neighborhood are
totally quiet. It makes a faint low-pitched whirring sound that I
guess might be 20-22 dBA; I have no way to measure it. The reported CPU
temperature varies from 50°C when idle, to 75°C running two copies of CPUBurn.
As mentioned before, the 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.
OVERCLOCKING REVISITED
Since the fan rearrangements and ducting improved cooling so much, I decided to see if the clocks could be turned up even higher.
As before, I started with the memory. I was able to run memtest86+
at 683 MHz with the DRAM/CPU/MCH
voltages set to 2.0/1.30/1.60 and the latencies set to 4-4-4-12-4. Note
that I increased the MCH voltage and lowered the DRAM voltage from
before; this eliminated occasional Prime95 failures. Next, I tried
cranking up the FSB. This required increasing the CPU voltage another
notch to 1.325V, to keep the measured Vcore above
1.2V running CPUBurn. With this voltage, the CPU was stable
with the FSB running at 240 MHz. Last, I sped up the graphics card
another 2%.
I ended up with the FSB/DRAM set to 240/640 MHz,
and the GPU/GRAM set to 427/1110 MHz, with no failures or instabilities.
CPU-Z reports the CPU clock as 3.607 GHz.
PERFORMANCE MEASUREMENTS
Here is a table of system parameters, memtest86+, 3DMark05, and PCMark04 benchmark results for the stock system, the
October configuration, and the final configuration.
| Parameter |
Factory Settings |
October 2005 |
March 2006 |
| FSB, DRAM, CPU clocks |
200, 533, 3.00 |
230, 613, 3.45 |
240, 640, 3.60 |
| Graphics GPU, GRAM clocks |
400, 988 |
420, 1098 |
427, 1110 |
| DRAM, CPU, MCH voltages |
1.80, 1.35, 1.50 |
2.10, 1.30, 1.55 |
2.00, 1.325, 1.60 |
| DRAM latencies |
4-5-5-15-4 |
4-4-4-12-4 |
4-4-4-12-4 |
| Memtest86+ L1, L2, memory MB/s |
21054, 18470, 3016 |
24174, 21208, 3463 |
25224, 22129, 3614 |
| 3DMark05 |
4903 |
5292 |
5379 |
| PCMark04 CPU, memory, graphics, disk |
5931, 4752, 6184, 4262 |
6859, 5445, 6690, 4302 |
7139, 5693, 6795, 4441 |
| PCMark04 System |
6060 |
6966 |
7261 |
CONCLUSION
This was quite the project/hobby/obsession. I learned a lot, and exceeded my goals. The resulting system is very fast, almost totally silent, and
finally done (yeah, right!). I look forward to actually using it rather than working
on it. That should be "tons of fun".
* * *
Editor's Note: Much thanks to Chris Thomson for sharing his experience with us.
Chris Thomson's moniker in the SPCR forum is cmthomson, and he may be reached by email at cmthomson at comcast dot net.
* * *
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
* * *
Discuss this article in the SPCR Forums.
| Help support this site, buy from one of our affiliate retailers! |
|
