Our Lapped CPU Heatsink Test Platform

Cooling | Reference|Recommended
Viewing page 1 of 4 pages. 1 2 3 4 Next

Anyone who's been reading SPCR for even a short portion of our 10+ year history knows that we take consistency and repeatability seriously. Our vision for SPCR is a database library for computer gear acoustics and thermal behaviour consistent enough so that you can compare something reviewed today with one in the same class of gear reviewed years ago. This usually means using the exact same base components in a test platform, using the same high precision measurement tools (whether it be for degrees, decibels, watts, grams, millimeters or millivolts). We've worked hard to stay true to this vision.

For testing CPU heatsinks, we've maintained from the start that a real CPU and a real motherboard are are necessary, they cannot be replaced with any simulations, no matter how much more convenient they may be. Not only does a CPU covered with an Integrated Heatspreader (IHS) simply not behave the same way thermally as a CPU-shaped copper plate heated by an electric coil beneath it, the heatsink mounting hardware can rarely be attached the same way as on a motherboard. The mounting system and its ease of use, its ability to apply high, even pressure to the CPU/heatsink interface are critical. This aspect of heatsink testing usually gets shortchanged when a non-CPU/motherboard test platform is used.


But as CPUs, sockets and motherboards came and went through their increasingly shorter life cycles, we've had to upgrade our test platform periodically. This means there are a few unavoidable discontinuities in our database. Besides, there's little point comparing, for example, a heatsink for a 65W P4 on a 775 socket against a heatsink for a 130W i7 on a LGA2011 socket: They are completely incompatible.

The last big change in our CPU heatsink test platform was a switch in January 2010 from the previous 95W TDP Intel Pentium D950 on anAsus P5Q-EM board to a 130W TDP 3.2 GHz Core i7-965 Extreme with an Asus P6T SE motherboard. There were two main reasons for the change:

The D950 was not challenging enough any more: The D950 CPU simply didn't run as hot as the latest socket 1366 i7 processors that had become the new performance kings at the time. The old platform could not differentiate enough between the high performance heatsink models designed to take on processors that run 35W hotter. A hotter 130W 1366-socket processor would help differentiate the A+ rank from the A and A- performers.

No more replacement parts, mainly, the motherboard of the old platform: A motherboard used as a CPU heatsink testing platform takes a huge amount of abuse. Intel informed me years ago that CPU sockets are designed for maybe 20 reliable uses. Oh, it will generally survive beyond that number of CPU insertions and removals, but the point is that it's reasonable for motherboard and motherboard component makers to build them for a moderate amount of handling. CPUs, motherboards and heatsinks are components that usually get seriously handled only a few times in their lifespan, during installation, a CPU upgrade, or maybe a heatsink upgrade. But our CPU heatsink test platforms get handled — and manhandled — every time we test a heatsink. The CPU gets pulled off and put back on at least a couple times, and the heatsink is often installed and removed as many as half a dozen times in our examination of the installation process. In fact, the Asus P5Q-EM that was our test platform in 2010 was actually the second sample of that model used with the D950 CPU for heatsink testing; we'd already broken the first one. We would not be able to get a third sample; more unused samples of that model simply could not be found any more. This was another good reason for a change.

The LGA1366 i7-965E Cooler Test Platform


Since the 3.2 GHz Core i7-965 Extreme CPU and Asus P6T SE motherboard came into service (nearly four years ago now), we've tested over 40 heatsinks on this platform. This is not all of the heatsinks we've reviewed, only the larger coolers designed for top performance. Some 20 or so less ambitious models were tested on a cooler-running AMD test platform, then later, on a 95W TDP Intel socket 1155 platform that replaced it.

We — and presumably most of our audience — have been reasonably pleased with the performance of our 1366 test platform. The newer LGA2011 socket has not seen a further increase in CPU heat; the TDP of all LGA2011 CPUs remains at 130W, which keeps the thermal core of our test platform perfectly relevant. Admittedly, one socket 2011 model, the i7-3970X, is rated for 150W, but this $1000 item is an outlier; CPU TDP continues trending downwards. Even the heatsink mounting mechanism for the LGA2011 hasn't really changed much from LGA1366. Fundamentally, no one had any quibbles with our testing methodology or the accuracy and relevance of our results. Of course, no one tests CPU coolers at the extremely low noise levels that we do routinely.

Still, a few heatsink sample suppliers have been less satisfied, and they've implied once or twice that SPCR could be biased for some brands. One of the more public of these incidents of supplier dissatisfaction occured a year ago with the Dark Rock 2 Tower Heatsink, three samples of which were submitted by be quiet! All three samples exhibited mediocre performance, especially given the large size and surface area of the fins. All three had poor contact footprints between their bases and the CPU, which seemed largely responsible for the poor performance.

The bequiet! representative questioned whether our testing method could be flawed, because, apparently, other reviewers had achieved good results. He even asked whether our use of voltage control instead of PWM control for the cooling fan could have played a part in the poor results. In an effort to dispel any hint of foul play or error on our part, I took extra steps to examine and re-test the three samples in great detail, adding an extensive postscript that conclusively showed the poor contact between the Dark Rock 2 Tower bases and the top of our test LGA1366 CPU.

That postscript contains two key photos and captions which I reproduce here:

Imprint of Thermal Interface Material on CPU and base of Dark Rock 2. A line on the top and bottom edges of the heatsink, matched on the CPU, shows where the TIM has been pressed away. Yet, the TIM in the center still has those fractal-like striations. This is conclusive evidence that the base of the bequiet! Dark Pro 2 sample here is concave.

Desired TIM imprint with the same CPU/motherboard, and a Prolimatech Megahalems heatsink. The heatsink was turned 90 degrees for the above photo, and the TIM was smudged a bit during removal, but you can clearly see the matching imprints of the TIM on the base and on the CPU. Both have a central area where there is very little TIM; it's been pushed from the center due to very tight contact. Only around the periphery are there any striations of the TIM. This cooler performed a whopping 12°C better than the Dark Rock 2.

These images are reproduced because they show how and why the profile of the heatsink base can have a dramatic effect on CPU cooling performance. The left photo shows that the contact between the DP2 base and the CPU surface was actually tightest along the perimeter of the CPU rather than at the center, where there was an obvious gap and no metal-to-metal contact. Since the center is where the CPU die resides, this is where it is hottest, and where the bond between heatsink and IHS is critical.

1 2 3 4 Next

Reference|Recommended - Article Index
Help support this site, buy from one of our affiliate retailers!