Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors Home PC Hardware from Made-in-China.com Search: Enter your search terms   Web www.lostcircuits.com Submit search form	.
Prices:

In our last heatsink round-up, we were looking at some reduced noise coolers that would be capable of handling the extreme heat generated by an AMD Athlon overclocked to 1715 MHz (roughly equivalent to a Quantispeed rating of 2100 MHz) with a core voltage of 1.84 V. Whoever read the review might remember that there were only very few candidates that actually qualified to handle this beast without running into heat-induced lockups and BSODs under full load. We have since been asked to take a look at some more coolers potentially better equipped to remove thermal load from the CPU. Among those brave enough to accept the LostCircuits challenge were Gabe Rouchon (Swiftech) and Toni from Plycon both of whom supplied us with about the finest coolers ever made by mankind (hehehe).

• - ASUS A7V266
• - AMD Athlon XP1900+ at 1715 MHz (12 x 143 MHz; 1.84V)
• - ASUS V8200 GeForce3 Ti500
• - 1 x 256 MB Muskin High performance
• - IBM DTLA 307020 20 GB HDD
• - Pioneer 12 x DVDROM
• - 3Com905c 10/100T Ethernet adapter
• - Windows 98 (First Edition)

Some Basics (safe to skip if you're not intersted in thermodynamics)

Before I go into the details, here is a quick run-down of what is involved:

• - Any CPU is a dynamic source of heat where the heat generation not only changes on a temporal scale but also involves very pronounced spatial separation between hot and cold areas on the die itself. Those latter hot / cold spots are caused by the die layout: cache, FPU and ALU units are not equally stressed by any given application.
• - Examples are gaming applications that challenge foremost the FPU units whereas the cache is only marginally involved or, conversely, benchmarks like ZDNet Winstone ContentCreation that involve the cache as well as the ALU units on the die.
• - It is trivial that the different areas of activity result in different patterns of heat across the die.
• - Different patterns can cancel each other out but also lead to hot spots on the die, however, under normal circumstances, that is, with adequate cooling, it will not matter because the dissipation of heat to the cooler will be fast enough to even out any differences.

• - Dynamic changes of heat generation and heat dissipation are an extremely important factor that need to be taken into consideration when looking at heatsinks.
• - The die itself and its packaging act as insulators, meaning that with increasing heat production, relatively less heat will reach the surface unless there is constant activity. If this were not the case, clock speed, power consumption, heat generation and heat dissipation would be on a linear scale. As it turns out, clock speed, power consumption and heat generation are directly proportional, heat dissipation, on the other hand is hampered by insulation effects of the die itself and the package unless there is a constant load on the CPU.
• - For any (or all) of these reasons, I believe that mimicking a CPU with a static peltier element to evaluate the performance of a heatsink is a valid approach but misses out on some very important issues.

• - First, there is the issue of overall cooling capability. Briefly, a lower thermal capacitance and high thermal conductivity, will absorb the heat fast and move it away from its source to release it into the environment. This is facilitated by a greater temperature delta.
• - This, however, bears the inherent problem that low thermal capacitance is not going to be able to handle heat spikes and consequently, the system will crash. One example is this CopperPlug that consistently gave the lowest temperatures on a PIII, but wouldn't allow any AMD Thunderbird to even POST successfully.
• - Second, there is the issue of thermal capacitance vs. cooling. My favorite example is the needle and the bowling ball. A needle heats up fast but cools down fast as well as soon as the heat source is being removed. A bowling ball takes awhile to get hot but once it reached ambient temperature, there is no turning back.
• - Guess which one is the more stable cooling device even though it might run way hotter under idle conditions?

All in all, the essence of this short excursion is to explain why one cooler, even though it provides lower idle temperatures can be much less efficient under load to the point where the system crashes because of exceeding the maximum tolerable operating temperature. Spinning this a bit farther, the optimal cooler would be a compromise between heat buffering (high thermal capacitance) and rapid dissipation (converting thermal load into fast dissipation of energy). The result is the concept we have seen emerging in the past few months: A bi-metal device composed of a heavy-metal core embedded into aluminum with low thermal capacitance, which, therefore, gives off the energy rapidly into the environment (air). The trick in this case is that the limited heat transfer through a relatively small die surface area to the core is getting turbocharged by what could best be described as sucking out the energy through a large surface interface by the surrounding aluminum.

Noise, Laminar Air Flow and Turbulence

The exchange of energy (heat) between the heatsink and the air being pumped through the fins depends on the surface area. Surface area can be increased by adding structure, however, there is a limitation to how much structure can do. For me, the best example goes back to my own days of building windsurfers. The paradox was that a smooth board was slower than a board with sandpaper-like surface. The explanation is that a micro-textured surface can cause a distinct interface layer where individual molecules of air (or, in this case, water) are trapped which act like ball bearings and prevent any interchange between the board and the surrounding media. This phenomenon is the reason why polished fins can have better heat exchange characteristics than porous ones with a nominally higher surface area. In other words, if surface is to be increased, one better make sure that the structures are enough to create surface increase without leading to pockets. There are other considerations like laminar flow, flow resistance and back pressure that influence the performance of a heatsink but it would lead too far to go into all the details here.

Dead Center or Eye of the Storm

Conventional cooling fans have one big problem. The area that is supposed to receive most air flow is the center of the heatsink, yet, the motor of the fan in its center actually blocks the air flow right where it would be most important. Consequently, the periphery of the cooler receives most air and will be cooled better than the center that is in direct contact with the CPU die. The possible work around is to not use an axial but a radial fan also known as squirrel-age blower. A side effect of these blowers is lower noise. The first widely publicized description of this concept goes back to Brett "3fingers" Jacobs' V2 setup: "Chilling" is probably the best description of this rig.

next page:    => The Heat Is On=>