Quantcast
 
 


2015 CPU Cooler Testing Methodology

Author: SKYMTL / AkG
Date: December 30, 2014
Share |

In the last few years the CPU cooler market has seen some drastic changes and many of those have focused upon delivering more user control over temperature and acoustics. There’s a real push for that optimal balance of low noise and acceptable temperatures but many also want the ability to boost performance when overclocking or when a particularly demanding application pushes the system beyond typical loads.

Actually delivering adaptable performance from a cooler isn’t all that hard but in some areas it proved to be a challenge. People using air coolers have always found ways to modify their heatsinks’ characteristics via resistors (Noctua’s LNA and ULNA Noise Adapters are good examples of this) or by adding higher performance fans. All in One water coolers on the other hand typically have their fans tied at the hip to the water block’s controller or to the motherboard’s fan headers so they presented the same set of challenges.

That was the situation when we first set up our CPU cooler testing methodology but, as technology has advanced, things have changed at a pretty drastic pace. Now water cooling manufacturers are including either hardware or software-based controllers that allow for a nearly infinite number of fan speeds with their mid-tier and higher end solutions. Even the air cooling market can now rely upon motherboard BIOSes which include increasingly fine-grain adjustments over rotational speeds and custom fan speed curves which increase and decrease airflow (and noise) based upon system load and CPU thermal output.

In our previous methodology, we used a number of different ways to simulate various fan states. Our standard tests utilized a combination of the coolers’ stock fans, 1300RPM Notcua NF-P12’s and 1900RPM Scythe S-Flexes for a broad, apples to apples comparison between units. Unfortunately, as we alluded to, that only scratched the surface of the possibilities as they exist today.

Another thing that we were a bit behind on was CPU technology. While the i7-920 we used was a beast in its heyday and the amount of heat it put out was copious in the extreme, Intel’s 3D transistor technology changed the cooling ballgame in a big way. By stacking transistors on top of one another, Intel cut down on die space but also focused the heat-emitting processor components into one densely packed area. The CPU’s own internal heatsink does dissipate the excess thermals to a certain extent but not all that well. As such, older coolers that performed well on pre-Ivy Bridge architectures tended to be poorly prepared for the new challenge and became overwhelmed by newer CPUs, particularly when the processors were overclocked.


How We’ve Moved Forward


With all of these things taken into account, we’ve (finally) modified our testing methodology to better reflect the market’s current realities. This has involved three basic areas: the test system, how the heatsinks / water coolers are tested against one another and how the results are shown.


Test System & Environment

Beginning with the test system, we’ve decided to go with an Intel Haswell i7-4770K processor which is benchmarked at three different speeds with all of its eight threads enabled: at stock, at 4.0GHz and finally at 4.5GHz. The tests are done on an open test system so the ambient temperature (22°C) can be effectively controlled throughout the testing process. Our previous tests were conducted in a closed chassis and ambient temperatures proved to be nearly impossible to replicate from one test to the next.

We also continue to use Arctic MX-2 thermal paste which is applied after thoroughly cleaning the cooler’s contact plate. It was allowed to cure for 24 hours under moderate to high loads (with periods of low loads) prior to testing, reflecting the realities of longer term use.

Loading conditions were replicated using Prime 95’s Small FFT test across all threads and monitoring is done with the RealTemp plugin for RiveTuner. Temperatures seen in our results are the peak readings after 15 minutes of full load.

The rest of the test system respects our intent to use passively cooled or near-silent components so outside factors won’t impact acoustical results.

Motherboard: ASRock Z97 Extreme6
Memory: 6GB Mushkin Silverline Stiletto DDR3-1600
Graphics card: EVGA GeForce GT 240
Hard Drive: 1x 240GB Intel 520 SSD
Power Supply: Topower Powerbird 900W


Looking Closer at the New CPU Cooling Charts

New technologies have allowed for a crazy amount of user control over fan speeds and general cooling performance versus acoustics but our older bar charts couldn’t show how these elements affected CPU temperatures. While different fans were used in an effort to represent changing conditions, they gave a very narrow view of things relative to the fan control capabilities of newer coolers and motherboards.

So how was this changed? We’re now testing all CPU coolers with their stock fans but modify fan speeds via either built-in software (if it’s compatible) like Corsair’s LINK or NZXT’s CAM or the motherboard’s onboard controller. As a result, our charts have undergone a significant facelift which represent how each cooler performs at a given speed.


So as not to overwhelm you with a deluge of results on an utterly confusing chart, we’ve chosen a quartet of very different all in one water coolers to compare against one another with the processor overclocked to 4.5GHz. Two stock fans are installed on each cooler in this example, even if some of the units only come with a single fan. In addition, the chart stops at 100°C since after this point stability and processor longevity become very serious concerns. Anything above that point is considered an absolute failure.

One thing to remember is that due to input voltage fluctuations from one fan to another, there could be a slight rotational speed difference (typically 5-10%). We’ve also decided to start our charts at 1000RPMs since this seems to be the bare minimum for achieving stability on an i7-4770K with any heatsink.

Looking at the results themselves, you will notice the coolers end their performance curves at different RPM locations; the H80i at 2750RPMs, the H55 at 1500RPMs, the TD03 at 2500RPMs and X41 at 2000RPMs. This simply represents the maximum speeds of their respective fans. The Silverstone cooler is a lone oddity since it only begins its readouts at 1500RPMs, the minimum speed at which the fan would operate before failing to spin up without outside help. Finally, even with two fans powering it along, that Corsair H60 fails to achieve sub-100°C temperatures at lower RPM settings as its results simply fly off our chartable area. It clearly isn’t capable of adequately cooling an overclocked Haswell processor

In the example above, NZXT’s X41 and TD03 reign supreme though the Silverstone cooler has an extra “gear” so to speak since it can operate at higher RPM levels. Their levels of cooling performance are better than the H80i while operating at significantly lower RPM levels.


The Acoustics chart follows in the same vein as the ones focused on performance but instead of separating dual and single fan results, they simply highlight each cooler in its default out-of-box configuration.

These tests are done in our open case setup with a sound meter positioned 30” away from the cooler and mounted on a tripod. The sound pressure meter used is a DT-805 which has been professionally calibrated and NIST certified while the results represent the highest levels obtained throughout the 20 minute load test.

Here we can see that two coolers come stock with a dual fan configuration: the H80i and TD03 while the two others only have a single fan in their default configuration. Ironically, having two fans isn’t necessarily a detriment since the TD03 is actually quieter than the H55 and X41 between 1500RPMs to 2000RPMs. This naturally changes as its speeds increase but neither of those other AIOs has the capability to offer higher RPM ranges for those who want to sacrifice noise for raw performance. The H80i meanwhile is one loud cooler while the H55 is a middle of the pack option in the noise department.


This ends the updates to our CPU cooler testing methodology. While we didn’t include any of them in the charts, the air coolers will be treated the exact same way as the AIO’s above.

Hopefully these changes to our testing practices allow for a broader understanding of what this new generation of coolers is capable of. Since they can easily go beyond a cookie-cutter format, so should our reviews move beyond an overly simplified approach to testing and presenting results. This should allow buyers to make more informed decisions since every option available in today’s market has the ability to be as quiet or loud as they could possibly want. However, what will determine overall appeal is how each individual approach to acoustics ultimately affects performance.
 
 

Latest Reviews in Cooling
May 10, 2017
All in one water CPU coolers are everywhere these days but Fractal Design's new Celsius S36 dares to do things differently. At 360mm, it may be long but its price to performance is incredible....
August 11, 2016
Prebuilt AIO or a potentially complicated custom loop? Alphacool believes we shouldn't have to choose and their Eisbaer 240 combines both under one roof and could blaze a new trail....
February 11, 2016
$130. Would you pay that amount of money for a CPU air cooler? Not many would but after seeing Cooler Master's MasterAir Maker 8 in action you may change your mind....