散热新主张：无风扇CPU热管散热系统

October 5, 2003 by Fred Mah

This fanless CPU cooling project is an exercise in logical design and simple execution using available technologies. It is far from simplistic. Fred Mah's fanless heatpipe cooled CPU system shows us a new approach to integrated system design that opens up all kinds of interesting possibilities. The cooling power of his system is nothing short of impressive. His project is a valuable contribution to the SPCR information "coffers".

Fred, btw, is a mechanical engineer who designs and manufactures various types of products and components, mainly for the audio industry. This probably helps to explain the resources accessed to have the various parts fabricated. Thank you, Fred! - Mike Chin, Editor / Publisher

* * *

Much effort has been made in recent years to minimize noise generated by CPU cooling fans, a fact that has been demonstrated by the popularity of variable and low speed fans coupled with efficient CPU heatsink designs. Even with the adjustable fans generating lower noise at lower speeds, the main noise sources in a computer system are fans and hard drives. Therefore, the best way to eliminate the noise is to remove these sources. As it is impractical to get rid of the hard drives, it seems like a good idea to cool the CPU without a fan. After looking at products based on heatpipe technology, such as Zalman's graphics card coolers, I felt it would be a good idea to try passive CPU cooling utilizing heatpipes.

HEATPIPES

Heatpipes are capable of transferring a large amount of heat per a given volume of working fluid due to the phase change that takes place. Inside of a heatpipe is a liquid under low pressure (or vacuum) that boils into vapor when it absorbs heat. This vapor then condenses back into liquid at the cooler surfaces of the heatpipe and releases the heat. So the concept here is to draw the heat from the CPU into one end of the heatpipe, while putting the other end of the heatpipe in contact with a larger heatsink to expel the heat into the air. (Editor's note: In layman's terms, it's sort of like watercooling without the pump. Here is a thorough, accessible explanation of heatpipes, by Thermacore.)

DESIGN

After studying what was done with other passive CPU cooling projects by experimenters like bluehat1 and numano3, I determined that placing the heatsink on the opposite side of the motherboard would be the best configuration:

• This would allow the creation of a case, or modification of an existing case, that could support the motherboard and heatsink in a way that wouldn't obstruct the motherboard components.

• It would also not limit the size of the heatsink that could be used.

• This was also optimal for placement of the heatpipes, since they would be used as the bridge between the CPU and the large heatsink.

• Additionally, a good amount of heat would be carried to the outside of a case design with this configuration.

PARTS

At the beginning of the project, I purchased the following components for the heart of the system, chosen mainly on the basis of cost:

-- AMD Athlon 1700+ (Thoroughbred) - about $50

-- Shuttle AN35N Ultra motherboard: The least expensive nVidia nForce2 motherboard with the four heatsink mounting holes that I came across. It also had passive chipset cooling and cost around $80.

-- Heatsink: I happened to have a sample piece of aluminum heatsink extrusion that seemed large enough for the passive cooling job. The dimensions are 9 7/8" wide by 12" long by 1 5/16" tall. In order to see if this would do the job, I had to look at the range of heat that the CPU would be generating.

From the highly useful Processor Electrical Spec web page by Chris Hare, it can be seen that AMD's Athlon XP family of processors go up to 2.2 GHz in the 3200+ model which generates approximately 80 W of heat maximum. The 1700+ on hand dissipates about 50 W maximum.

The thermal resistance for natural convection of the heatsink was known to be 0.91 °C/W for a 3 inch piece. According to a technical document at Wakefield Engineering's website, the thermal resistance would decrease by half if the length of the part were increased by four times. Therefore, the 12 inch long extrusion should have a natural convection thermal resistance of 0.455 °C/W. In an ideal situation with no thermal losses between the CPU and the heatsink, this would mean that dissipating 80 W of heat would leave the CPU at a temperature of 36.4°C higher than ambient. So for a 25 °C (77 °F) room temperature, this would be a CPU temperature of 61.4 °C (142.5 °F). The maximum allowable CPU die temperature for a 3200+ Athlon is 85 °C, so the thermal resistance of this heatsink is theoretically sufficient for any Althlon XP processor available at this time.

A design concept was drawn up in Solidworks, and from this the dimensions for the heatpipes, copper blocks, large heatsink mounting holes, and test rig were determined. The idea was to have a minimalist test rig to quickly build and test the performance of this cooling system. I had the sheet metal stand portion fabricated and spray painted it with white primer to prevent rust.

Solidworks drawing of the proposed test rig

The pipes were not snug, but not too loose either, and after tightening the screws they were pretty well fixed in place. The heatpipes were a little shorter than I was expecting, so they don't quite stick out of the top of the copper block, as seen in the pictures below. Before tightening the blocks down, Artic Silver 3 was applied at the interface between the CPU die and the large copper block. Then generic silicon grease was put on all other contact surfaces. I didn't use AS3 since there was a fair amount of area to cover, and it would have used a lot of the AS3.

THE SYSTEM

Test fitting the parts

Close up of blocks and heatpipes

Large heatsink mounted on backside

FIRST TEST

After all the components were secured in place, I fired up the system and went into BIOS. Initially I used a two-fan Antec Smartpower power supply that I happened to have available. Later I put in a Seasonic Super Tornado 300 (SS-300FB). The Seasonic is much quieter.

When I turned on the system without a hard drive, I put my head a few inches from the motherboard to see what I would hear. I didn't really hear much but the power supply fan. Ocassionally the motherboard generated various kinds of electrical hum and crackles depending on the load, but this is likely to vary with the motherboard design and CPU load. Such electrical noise was mostly apparent when overclocking the CPU.

I sat in the BIOS screen with the temperature readings. I was happy to see the temperature did not ramp up quickly, it slowly went to a stable reading after many minutes.

I spent a fair amount of time trying different CPU speeds with various overclocking settings. This was done to simulate different CPU heat loads and see whether the CPU temperature would go out of control. The highest speed Althon XP is a 3200+ at 2.2 GHz. I was able to run the system for around 10-15 minutes at that speed, but the system was unstable, because I received a system error in Windows. The temperature was in the low 60s °C when the error occurred. I suspect the CPU/RAM combination is the main factor in the overclock success/failure, not overheating. The system seemed to be fairly stable at 2 GHz. Perhaps PC3200 RAM instead of PC2100 Kingston "value ram" would make a difference in the maximum overclocking tests.

TESTING & RESULTS

The powered on system

I performed a clean installation of MS Windows 2000. I used Speedfan 4.08 to monitor temperatures, and CPUBurn with Runprio with the following command line "runprio -x high burnk7" to perform a AMD K7 class burn at high priority. This setting provides at least 5°C higher temperatures than Prime95. The Shuttle motherboard has no thermal probe in the CPU socket, so the readings from Speedfan were assumed to be from the onboard thermal diode in the CPU. I did not make any thermocouple measurements since I don't have that kind of equipment yet. The temperatures at load were taken after CPUBurn had been running more than one hour. Room temperature was taken with a digital multimeter with temperature probe that was placed below the large heatsink.

Idle Temperatures
Athlon XP	Setting	Room °C	CPU °C
1.46 GHz	1700+ stock: 133 MHz FSB x 11 multiplier, 1.60V	23	40
1.46 GHz	1700+ undervolted: 133 MHz FSB x 11 multiplier, 1.25V	23	37
2 GHz	1700+ overclocked: 200 MHz FSB x 10 multiplier, 1.65 V	22	44

*

Temps with CPUBurn: Full Load
Athlon XP	Setting	Room Temp °C	CPU Temp °C	Temp Rise °C	CPU Power (W)	Thermal Resistance (°C/W)
1.46 GHz	1700+ stock: 133 MHz FSB x 11, 1.60V	22	52	30	49.4*	0.61
1.46 GHz	1700+ undervolted: 133 MHz FSB x 11, 1.25V	22	44	22	30**	0.73
2 GHz	Maximum overclock: 200 MHz FSB x 10, 1.65 V	23	64	41	68.3~71.6**	0.60~0.57

* From Processor Electrical Spec web page by Chris Hare
** See Calculating CPU Power & Thermal Resistance at bottom of page for details.

The system works well enough to run over a wide range of Athlon XP CPU speeds, keeping the CPU temperature below the 85°C maximum. This is a very positive result. Even the current fastest XP, the Athlon XP-3200+ w/Barton core, can probably be cooled well enough with this setup: Its heat dissipation of 76.8W would mean a max temp rise of 47C. In a 22-23°C room, the max CPU temp would then be ~70°C -- still well below the 85 °C max.

(Editor's note: The 41°C temperature rise with a 68.3W CPU is obviously higher than the theoretical calculation, presented earlier in the article, of a 36.4°C rise for a 80W heat source cooled by this heatsink. This is due to unavoidable losses in the heat transfer between the CPU and the heatsink.)

The noise level of the CPU cooling system is virtually nil; the only significant noise in the system comes from the the power supply and hard drive. As expected, the hard drive is the biggest offender, and I often turned the system on and off without the hard drive connected just to "hear" the system.

ANALYSIS & CONCLUSIONS

The only part of the cooling system that became too hot to touch sometimes were the copper blocks mounted on the CPU. During the load tests at 2 GHz, these could be touched for a few seconds, while the rest of the components could be touched for longer periods of time. The copper blocks at the cooler end of the heatpipes were able to be touched during the testing, and were noted to be somewhat cooler than the hotter end. The large heatsink was always warm, but a hand could be placed on them almost all the time.

I also tried aiming a floor fan at the large heatsink to see what effect this would have on the CPU temperatures. During a load testing, the CPU temperature was approximately 5°C lower than without the forced air blowing on the heatsink. This was a rough estimate to gauge the performance of the heatpipes. If the temperature had not dropped, then this would have indicated that the heatpipes were a limiting factor in cooling. Therefore, the temperature could be even lower if design changes were made. These could include increasing the surface area of contact onto the large heatsink, improving surface finish smoothness and fit, using only the highest performance thermal compound, making a vertical tunnel for airflow over the heatsink, and increasing the size of the large heatsink. There are many ways to modify this system, but the existing system has proven to function very effectively at both reducing noise and cooling the CPU in a passive manner.

This design works well for normal usage, and most likely even for extended gaming sessions. The only things necessary for a complete quiet system would be to silence the hard drive, use a passively cooled video card, and a very quiet or passively cooled power supply. Use you own imagination about how to physically integrate drives and a power supply to this system.

---

** Calculating CPU Power & Thermal Resistance

1. For the 1.25Vcore at 1.47 GHz: It is known that at 1.6V and 1.47 GHz, the T-bred core XP dissipates 49.4W. Since CPU wattage is directly proportionate to Vcore:

• 1.6V squared = 2.56
• 1.25V squared = 1.5625

The latter figure is 39% lower that the former. Thus at 1.25Vcore, the dissipated power is 61% of 49.4W or ~30W. This result jibes with Kostik's nifty CPU Power calculator utility

2. For the 2 GHz overclocked state: According to the Processor Electrical Specs web page, the Athlon XP-2400+ Thoroughbred is clocked at 2 GHz at 1.65V (the same settings as this overclocked XP1700+) and dissipates 68.3W. However, plugging in those settings for a XP1700+ in Kostik's CPU Power calculator utility yields the higher 71.6W figure. So both numbers are provided.

In any case, as Russ (Rusty075 in the SPCR forums) pointed out, the changes in thermal resistance with a hotter or cooler CPU are to be expected: As the temp of the heatsink goes up, the speed of the air moving across the HS increases, thanks to the "stack" effect. That will drop the °C/W - hence the <0.6°C/W at the maximum overclock ~70W setting. The opposite effect occurs with a cooler CPU -- the airflow decreases, reducing the cooling effect. Hence the 0.73°C/W at the 1.25 Vcore 30W setting.