Testing & Results
The CPU coolers tested were installed in a computer case in its normal, upright orientation (a NZXT H630). A 200mm top/rear exhaust fan was added to the enclosure to aid in cooling VRMs and most of the front drive cages were removed to clear the path from the 200mm intake fan. The GPU remained installed during testing. All fans were set to 100% to remove that variable from the results (motherboard fan control was disabled). In the case of liquid coolers, they were mounted in the rear exhaust location if possible (otherwise, as in the case of the Swiftech H220 and the TD02*, mounted in the “floor” of the H630 as an intake). This is how I would assume most enthusiasts would set up a similar case while overclocking a similar platform.
All tests were performed using the AIDA64 Extreme Edition System Stability test, using 100% fan settings on an Asus M5A99FX PRO R2.0 (PWM/motherboard fan controls were disabled for testing). The test was allowed to run until temperatures plateaued, then I recorded the ambient temperature of the intake air and began logging temperatures over the next minute. After an initial warm-up run, I ran each test at least three times (more if I received inconsistent results), and recorded the ambient temperature again. Once I had “good data,” I dropped the best and worst results and subtracted the (average over the test) ambient temperature from the median result to arrive at the delta T temperature you see in the chart.
Each time a heatsink was swapped, the Tuniq TX-2 thermal interface material I used for each application was cleaned off of the contact surfaces with Arctic Silver’s ArctiClean two-step TIM remover, and an appropriate amount of TX-2 replaced for the next heatsink. Due to the nature of applying TIM and mating two surfaces, I would like to adopt a 3% margin of error – even though my thermometers and the built-in thermal diode measure temperatures down to one-tenth of a degree Celsius, it could be assumed that temperatures within a degree of each other are essentially the same result.
*The SilverStone TD02 was oriented and tested in the same location as the Swiftech H220, but the hoses were not long enough to actually mount the radiator to the chassis.
Motherboard: Asus M5A99FX-PRO R2.0 w/ 1708 BIOS/UEFI
System Memory: 8GB (2x4GB) GSkill Ares 1600MHz DDR3 CL8
Processor: AMD FX-8320 Piledriver, 4.6GHz/1.428V
Video: Sapphire Radeon 7950 3GB 1000MHz Core, 1300MHz mem
Disk Drive 1: OCZ Vertex 2 240GB
Enclosure: NZXT H630, +200mm exhaust fan (top/rear)
PSU: Rosewill Lightning 800W Modular 80+ Gold
Monitor: 1920×1080 120Hz
Operating System: Windows 7 Ultimate 64-bit w/SP1
I think I may need to reiterate what this chart shows (or rather, what it doesn’t). In order to test each cooler as objectively as possible, I turn off all fan controls and let them run at 100%. This chart shows the peak temperature that was reached during the stress test, but it does not show the volume or character of sound used to get that result. The tendency is to look at the chart and view it as a hierarchy, where the position on the chart determines how “good” a cooler is. While that is true for peak “all out” performance, there are many qualities other than performance that you may (or should) consider when choosing a CPU cooler.
Keep in mind an overclocked FX-8320 like this one will consume around 200W, depending on your specific components. That’s normally a substantial amount of heat for 120mm CPU coolers, and helps explain why a push/pull configuration with an additional fan on the NH-U12S gains an additional 2 degrees Celsius of cooling capability (hitting a max core temp of 33C over ambient).
While the Noctua NH-U12S fits snugly in between the Hyper 212 EVO and V8 GTS (a twin 140mm cooler, remember) it does so with an entirely different character. The noise generated by the Noctua fans at 100% is vastly different than the EVO’s stock fan and the twin 140mm POM bearing fans of the V8 GTS. Exactly how different is difficult to quantify, but I will say all of Noctua’s features are less of a bag of tricks and more serious engineering – all of those little details REALLY add up to a big difference in noise. The result is a product that performs similar to or better than other 120mm coolers without the typical noise penalty.