An English electronic engineer who likes making things, Bill Todd made many modifications, and created a new wood case to take this modest old Pentium III system far along the road to silent nirvana. His ingenious journey involved recycling and creating of all kinds of parts including old electronics heatsinks, home-made damping gel packs, scraps of plywood, and even wheels from an old scooter. Two years after he first assembled this passively cooled system, it’s still working silently away, even after a leak in a gel pack next to the hard drive.
Oct 25, 2006 by Bill Todd (btodd at cis dot co dot uk)
|An English electronic engineer who likes making things, Bill Todd made many modifications, and created a new wood case to take this modest old Pentium III system far along the road to silent nirvana. His ingenious journey involved recycling and creating of all kinds of parts including old electronics heatsinks, home-made damping gel packs, scraps of plywood, and even wheels from an old scooter. Two years after he first built this passively cooled system, it’s still working silently away, even after a leak in a gel pack next to the hard drive. |
– Mike Chin, Editor
The noise that most PCs make is very annoying and I have often wondered what it would take to make a truly quiet PC. When I found an old 1Ghz Pentium III motherboard (MSI 6315), a chassis, a spare HDD and an old micro ATX power supply, I thought I would try to make a truly silent PC.
The end result of my work on this P3 system.
It began life more like….
Given the relatively low power dissipation of the system, I decided to try and cool everything passively without any fans. This entailed a lot of customization and optimizing parts for convection cooling.
COOLING THE PROCESSOR
Normal Pentium CPU heat sinks tend to have closely spaced fins, which require a fan to force the air to move and remove the heat. The heatsinks used in the electronics industry, however, are usually rated for convection cooling (i.e. with the fins vertical in free air). This rating describes how hot the heatsink will get according to how much heat is dissipated by it, so the unit of thermal resistance is given in degrees per watt (°C/W). To calculate how big the heatsink needs to be, I would need to know the amount of power the 1GHz Pentium 3 uses (it’s safe to assume that all the electrical power the CPU uses will be dissipated as heat) and the safe maximum temperature for the CPU. Checking the Intel website, it gave figures of 29.9W and a temperature of 69°C. If I assume an ambient temperature (inside the case) of 30°C, then I can calculate I need less than…
…which is the thermal resistance between the CPU and the air.
This has to include (in my case) the thermal resistance of the processor to heat-shunt junction, heat-shunt to heatsink junction and the conductivity of the copper heat-shunt itself. The thermal resistance of the junctions is almost impossible to calculate, since it depends on the area of contact and thickness and conductivity of the jointing compound, but if care is taken to mate or flatten the contacting surfaces it is not too difficult to reduce them to less than 0.1°C/W.
Because the size of heatsink would be much larger than the socket, I decided to use a 40mm diameter copper block to act both as a spacer, lifting the heatsink clear of the motherboard components, and as a heat-shunt to move the heat from the CPU chip into the heatsink. The thermal resistance of the block is calculated from the formula T ÷ (l × A) where T = thickness, A= area and l = the thermal resistance of copper (385 W/m K). Plugging these numbers in, we see that the copper block barely adds 0.005°C/W to the thermal resistance, so I will need a heatsink better than 1.3°C/W.
I asked few friends to look in their junk pile and before too long Bill Hall gave me a large 120x120x60 mm black anodized heatsink with a rating of about 1°C/W. (Cheers Bill)
Oversized open-fin anodized heatsink on the CPU.
The copper block was lapped flat on a surface plate with fine grade abrasive paper then polished to a mirror finish on one side. The other side of the block was lapped to the heatsink with fine grinding paste and then polish to mate the surfaces exactly. I then mounted the block to the CPU socket using a stainless steel spring like a normal Pentium heatsink and fan. The aluminium heatsink was then mounted with a couple of M5 screws, to the chassis in such a way as to just rest on the copper block without unduly stressing the CPU. The screws go between the chassis motherboard tray and the part of the large heatsink that overhangs beyond the top edge of the motherboard. A small bracket on the bottom of the heatsink keeps it in place when the PC is moved.
To work well, passive or fanless cooling needs a clean airflow path, clear of obstructions. To this end I cut two large square holes in the bottom of the chassis, and matching holes in the outer wooden case, then cleared room above the CPU, by moving the CD-ROM drive as far forward (out of the case) as possible and, to clear space behind it, mounted the power supply vertically. The result is a CPU temperature that remains below 60°C (35 above ambient) even after a prolonged thrashing.
View from above CPU heatsink area; it’s open for convection.
View of heatsink from the underside of the case: Open again, for convection intake.
THE POWER SUPPLY
The power supply (PSU) is a 140W micro-ATX design, small but adequate to drive the motherboard, a network card, the hard-disk and CD-ROM.
I couldn’t find any figures for the efficiency of the PSU so, I measured the current draw on each of the supply rails (5V, 12V, 3.3V etc) with a DC clamp meter. Multiplying the current drawn by the voltage of each of the rails will give the power used by each rail. I estimated the power used at idle to be 28.5W. Measuring the mains input gave a power input of 46W, this equates to an efficiency of about 62%.
With the CPU running at 100% the whole machine is drawing about 60W which means I have about 22W (60W x 38%) of heat to dissipate in the PSU. I have not been able to find out which bits of the power supply dissipate the most heat but, clearly, the semiconductor parts are the most sensitive to temperature so, it makes sense to ensure that these part are adequately cooled.
It would be possible to remove the major semiconductors (i.e. switching transistor, diodes and regulators) from the PSU circuit board and mount and rewire them on a remote heat sink but, given the modest dissipation and small size of my PSU, I removed the fan and added heat shunts (made from short lengths of 4mm thick aluminium extrusion) between the semiconductor devices to the two heatsinks mounted over the fan orifice. I punched a few extra holes in the PSU case just to let the air flow through freely. (Editor’s Note: Bill could have used use a pair of needle nose pliers to twist the bars in the grill to 90 degrees to greatly decrease the obstruction.)
Modified PSU cooled passively with external heatsinks.
The PSU is mounted at 90 degrees to normal to allow the heatsinks to sit vertically and to clear space above the CPU heatsink for a clean air flow. This has the slight disadvantage that the power connector now protrudes from the top! The PSU runs quite warm rising to about 48°C at full power.
Note the extra holes drilled in the casing of the power supply to aid cooling.
WARNING! Readers should be aware that a power supply can have high DC voltages in its capacitors even after power is turned off. Heatsinks can also be “live” in some power supplies. Exercise great caution if you choose to open up any PC power supply.
SILENCING THE HARD DISK
I used an old 3.5″ IBM 20GB hard disk that had a reasonably low 6.5W power consumption, but an unreasonable sound level.
To silence the rather noisy hard-disk, I suspended it with foam rubber mounts inside a die-cast box (170mm x 115mm x 55mm) and sealed the lid with a 1mm diameter drive belt from an old VCR. The result was quite astonishing. The high frequency whine that the disk made became inaudible. The HDD case is itself rubber mounted to the chassis to remove most of the lower frequency vibration. Each of the four mounts for the HDD case consists of two 8mm M3 threaded spacers super-glued into a short length of 8mm diameter rubber tube.
HDD enclosed, gel-packed and suspended.
A 1mm diameter drive belt from an old VCR
used as the gasket seal for the HDD box.
Unfortunately the IBM HDD did not have the on-board temperature monitoring of some newer drives, so I could not monitor the temperature while from the desktop but, measurements with a thermocouple on the drive in its case showed it running at 35°C above ambient (52°C) too hot for comfort and too close to the manufacturers 55°C maximum.
I decided to try to reduce the HDD temperature by improving the thermal coupling between the HDD and its case. In order to avoid any mechanical coupling I decided to use a gel pack. Initially I looked at medical gel packs which are well made and sealed, unfortunately one I had was too large and I couldn’t find one advertised that was small enough to fit in the box with the drive, so I thought I’d try to make my own.
With a quick search on the web I found that water based gels are usually made with methyl cellulose and that this is found in many applications that need to absorb water, like disposable nappies (diapers) and, most usefully, wallpaper adhesive. I used a small zip-top bag, part filled with wallpaper paste mixed to the consistency of runny porridge, to fill the gap between the drive electronics and the lid of the die-cast box. The result was a 10ºC drop in temperature without any additional noise.
Original gel pack to help transfer HDD heat to the casing.
After a couple of months of use, I found that the foam rubber (actually polyurethane) had squashed down under the weight of the drive and, consequently, the drive had started to produce an audible hum. So, I took the opportunity to experiment further with the gel pack idea. I replaced the foam blocks, spacing the top of the drive away from the box, with a larger home made gel pack. After testing it for a couple of hours with some hard disk thrashing, the temperature of the drive reached 37ºC (about 15ºC above ambient). The drive was still pretty quiet (only the faint head seek clicks were audible in the workshop), but the large gel pack had coupled some of the drive noise to the HDD case.
V2: Larger gel pack.
Nine months later, I acquired a replacement for the original IBM drive, a much quieter 40GB Seagate ST-340810, and replaced the foam suspension with a rubber-based foam, used for sealing doubled-windows. The machine was then incredibly quiet; even the drive seek clicks were inaudible in normal use.
Version 3: A new quieter drive.
A NEW CASE
Having all but destroyed the original case while “improving” the airflow through it, I decided to make a wooden sleeve case to take the chassis. The case is made from a few off-cuts of birch ply that I had in the workshop (hence the size and weight!). A couple of pieces of 30mm thick for the sides and three pieces of 25mm thick for the top, front and bottom panels. The power and reset switches are made from a dome screw heads and a couple of M3 spacers. [Editor’s Note: Bill refers here to a very specific type of plywood, usually originating from the Baltic countries, that have many more “plies” than the usual plywood, made entirely from birch wood, and with minimal voids in the plies. It was the preferred material for English hi-fi speaker cabinets for its high density and rigidity before lower cost MDF wood sheets came along. You can see the unusually high number of layers at the plywood edges in the photo on the first page.]
The PC simply slides in through the open rear of the wood sleeve. A couple of wheels from an old scooter and a pair of casters make moving the heavy PC easier. Zinc plated perforated sheet wrapped over the top and forming a vent at the bottom of the case finished the job nicely.
The replacement Seagate drive improved this machine immensely, the drive head seek clicks are inaudible and the quiet hum produced by the old drive has disappeared completely. It is now quieter than the backlight inverter in my Viewsonic TFT LCD panel monitor (which squeals like a hungry mosquito). It’s certainly a much nicer experience to browse the web without the constant roar of a PC.
I’m not sure if it would be possible to passively cool a more power hungry CPU without resorting to exotic heat-pipes or water-cooling, the CPU heat sink would have to be too big to mount near the chip, but since many of the newer processors are far more energy efficient, it would be possible to use the same techniques to build a faster SilentPC.
After using the system for 19 months without opening it up, prompted by Mike Chin of SilentPCReview, I took the SilentPC back into the workshop to take some more pictures. While it was open, I decided to check the HDD Box for leaks.
Just as well…
On opening the HDD box I was presented by a sad sight: There were obvious signs of corrosion on the lid and a large amount of powder sitting on top of the small gel-pack. The stuff sitting on the gel bag is powdered ‘corrosion’ (aluminum and zinc oxides presumably, with some remains from the wall-paper paste). I suspect that the moisture/paste reacted chemically with the die-cast aluminum, quickly forming the oxide powder and never forming droplets.
A leak in one of the bags and corrosion but no damage to the drive.
Removing the drive, it’s clear that only the smaller single-bagged gel-pack has lost some water, the cause of the corrosion, although it hasn’t dried completely. The large double-bagged pack has survived quite well, no obvious sign of leaks.
The HDD itself seems fine. I suspect that, by corroding, the metal of box has absorbed the moisture lost by the gel pack. I’ll have to pack some silica gel moisture absorbing crystals into the box and double bag the smaller gel-pack before sealing it up this time.
* * *
Articles of Related Interest
* * *