Testing Methodology
HSF testing is not always a straightforward subject. I often find myself trying to account for every conceivable detail from interface material curing time to relative humidity in the room. Although many details are important, some are more sensitive than others are. We strive to keep our environment as consistent as possible while maintaining variables that are ‘realistically” controllable. Although our results will not be 100% accurate to the tenth of a degree, consistency will be maintained. Temperature diodes are rarely accurate to the degree and so long as the same testing methodology is maintained with all HSFs in our lab, a comparative analysis will be very valid. This is what we hope to achieve.
One can create very elaborate test benches for HSF testing. I’ve seen all sorts of load testing contraptions that are certainly very good at scientifically judging heatsinks. On the other hand, they do not provide a real-world view of what the consumer will see. There are so many different challenges HSF buyers have to face — backplates that don’t fit, component clearance issues, mounting issues etc. It is also nice to see how a heatsink fairs in a real PC, such hearing that “Heatsink X” kept a Q6600 at 59°C at full load for example.
Environmental Controls: Ambient temperature is perhaps one of the most critical control variables. As such, testing is conducted in a temperature-controlled environment. Ambient temperature is maintained at 21°C for testing purposes and is not permitted to deviate more than 0.5°C higher or lower than our target of 21°C. Every time we record a result, we also record the ambient temperature at the same time. You’ll notice that most of our tests are in the 20.8-21.2°C range. We do our utmost to keep this variable controlled. I prefer not to adjust results to reflect changes in ambient temperature. It is not fair to say that a 1°C increase in ambient temperature will always cause a 1°C increase in CPU core temperature. These types of things are not always so linear.
Hardware Configuration
Although most enthusiasts are interested in 775-based processors, there are many AM2/939 processors still in use today. We believe it is important and valuable to test each heatsink on both the AM2 and 775 platforms. Any given heatsink could have a great 775 mounting system but a very poor AM2 mounting system for example — we want our readers to know this. Below are the specifics:
AMD Test Platform
- AMD Athlon 64 X2 6000+ (AM2 Windsor Core, 3.0GHz, 2x1MB L2 Cache)
- DFI Infinity NF570-M2/G Mainboard
- 1x1024MB Buffalo FireStix PC2-6400 (5-5-5-18-2T)
- Maxtor 100GB SATA Hard Drive
- Pioneer DVR-212D SATA DVDRW
- Corsair VX550 Power Supply (41A +12V rail)
- Generic PCI graphics card
Intel Test Platform
- Lapped Intel Core 2 Quad Q6600 (G0 Revision, 2.4GHz, 2x4MB L2 Cache)
- Asus P5K-E Mainboard
- 1x1024MB Buffalo FireStix PC2-6400 (5-5-5-18-2T)
- Maxtor 100GB SATA Hard Drive
- Corsair VX550 Power Supply (41A +12V rail)
- Generic PCI graphics card
There is some logical reasoning behind our choice of components for testing. Firstly, we’re using two of the most popular CPUs from both AMD and Intel and coincidentally, two of the hottest. We debated whether or not to use an Intel dual core, but with the popularity of the Quad cores and common cooling challenges, it made more sense to stick with a Quad. Since the surface of the Q6600 we were using was far from flat, it was lapped to ensure a level surface. The Athlon X2 6000+, although not a quad core CPU, is still based on the 90nm Windsor core and can definitely get cooking — especially when increasing voltage.
The mainboards we chose are midrange boards and are not exotic. We tried to select mainboards without large onboard cooling contraptions to ensure compatibility with a wide variety of coolers. For memory, we decided to test with only a single DIMM. The reasoning behind this is that in some circumstances, very large coolers may obstruct some of the memory slots. We’d rather consistently use one stick as opposed to having to modify our standard platform in some situations.
The SATA drive used is of little importance, however we stuck with only a single drive to keep noise levels down and to prevent excess heat dissipation from unnecessary components. The Corsair VX550 PSU is a top-notch single +12V rail unit with plenty of power for a heavily overclocked Q6600. It is also a very quiet PSU that is suitable for a test rig. We decided to use an old, passively cooled PCI graphics card in order to keep noise levels to a minimum and to prevent an excess of heat from dumping in the case.
We install the small PCI graphics card away from the socket area so as not to introduce heat to the chipset, mainboard components and the CPU heatsink itself. One might argue that failing to use a “Common” gaming card like an 8800GTS for example is not representative of a high performance system — this is true to some extent. We would counter that argument stating that this is a CPU heatsink testing platform and that controlling temperature and acoustic variables is more important than being 100% representative of a common enthusiast grade system. Consistency is of utmost importance.
Our test bench is actually a stripped down Antec mid-tower case. It has a Scythe 46.5CFM 120mm exhaust fan mounted. The PSU fan also exhausts some heat from the CPU area. We decided to use an actual case to be able to check for ‘real world” clearance issues. We left the case on its side with the side panel open — we felt that this allowed easier control of the temperature within the case. The heatsink is allowed to “Breathe” as much as possible while still being in a real case of sorts. We had quite a few heatsinks to test, so easy access was another side benefit, of course!
Temperature Measurement and Load
Idle testing is performed by simply leaving the system at the windows desktop for 30 minutes, or until the temperature stabilizes. No applications are left running aside from required temperature/system monitoring applications. We decided to use three different load tests as opposed to just one. We noticed that there can be quite a difference in heat output depending on the type of load put on the processor — this was most evident on the Q6600. The first is multiple instances of Prime95, in-place large FFTs. The second is multiple instances of Prime95, small FFTs and the final test is multiple instances of a F@H Gromacs core.
All power saving features were disabled for testing. The processor “State” was watched carefully during testing to ensure that the CPU did not “Throttle.” Since our tests are quite demanding, we had to decide upon “Cut off” points to judge a heatsink “N/A” for a given test. We decided upon 85°C for our Q6600 and 75°C for our AMD X2 6000+. These are both about 10 degrees above the “Safe maximums” specified by Intel and AMD, so we give them a fair chance. Although we do take our test chips to hell and back, we do care about them and have to draw the line somewhere.
Temperature Measurements
Temperature measurement has become a very controversial topic lately, as mainboard sensors can be laughably inaccurate. New applications like CoreTemp pull temperature readings directly from the on-die sensors embedded within the CPU cores, which is ideal. Although I really like CoreTemp, others do not. Because of this, both the mainboard reported CPU temperature and the temperature of the hottest core according to CoreTemp will be included for all results. “Mainboard” temperatures will also be included for the Intel testing as the CPU fan can influence this temperature depending on the design of the heatsink. The DFI mainboard sensor did not seem to vary between heatsinks so “Mainboard” temperatures for AMD testing will be discarded. Please note that the temperatures reported by Smart Guardian on the DFI board are not accurate. They come in at least 7 or 8°C too low. At idle, temperatures are often reported below ambient. These measurements are very valid for comparative purposes but are definitely not accurate.
Monitoring applications
- CoreTemp 0.95.4 Beta
- Asus PC Probe II 1.04.19
- ITE Smart Guardian 2.03
Reference Fans
- 120mm: Scythe S-FLEX SFF21F.
- Rotational Speed:1600RPM
- Airflow: 63.7CFM
- Noise: 28dBA
- Size: 120x120x25mm
- 92mm: Thermalright branded 92mm fan
- Rotational Speed: 2400RPM
- Airflow: 45CFM
- Noise: 35dBA
- Size: 92x92x25mm
The Scythe S-FLEX has become somewhat of a staple item to enthusiasts and reviewers. Although we hate to be somewhat unoriginal with our selection of a 120mm reference fan, the S-FLEX really does provide the best balance between airflow and noise. At 63.7CFM and only 28dBA, it moves a fair bit of air and cannot be heard above other components in the average PC. Most “Quiet” fans are rated for a rather weak 40-50CFM. Another reason we chose the S-FLEX is that it has an open post mounting area, allowing clip mounting with many heatsinks including the Thermalright varieties. We have much higher output 120mm fans in the lab, however they are all 38mm thickness and are incompatible with many heatsinks. The mentioned 92mm fan is a personal favourite of mine. It was bundled with the Thermalright XP90 retail package that I bought a couple of years ago. It puts out a healthy 45CFM at 35dBA. It is not the most quiet 92mm fan out there, but does not compromise airflow for testing purposes. This is a 25mm thickness fan employing open post mounting, making it compatible with the vast majority of 92mm capable heatsinks.
Thermal Interface
Thermal interface material is an important factor when installing a heatsink. Unfortunately, some materials take upwards of 200 hours to achieve their maximum performance potential, like the popular Arctic Silver 5. We’ll be sticking with Arctic Ceramique as our standard thermal interface. It performs very similarly to AS5 and does not require the same very long cure time. Arctic Silver claims that Ceramique will perform only “slightly” better once it has completed its 25-hour “break-in”, whereas AS5 can realize significant improvements over the first 200 hours. Whatever improvement is realized during the first day of use with Ceramique is likely minimal enough to disregard which is why we chose this material.
Installation
The CPU heatspreader and heatsink base is always cleaned with 70% isopropyl alcohol prior to installation. Thermal interface material is applied per Arctic Silver’s Ceramique instructions available here: http://www.arcticsilver.com/ceramique_instructions.htm .
Base Testing
We conduct two tests to check for base flatness. The first is the “Glass pane” test. A small drop of water is placed on the base and a flat pane of glass is placed on top. If the base is not convex, the water will be evenly spread to all edges of the base. See below:
The second test is the “Straight edge” test. This test can be used to find a convex or concave base. A straight razor blade is placed on the base and held up to a light source; if the base is convex or concave, light will escape below the razor blade.
Overclocking
Several different overclocking “Reference points” were selected on our test systems to determine how well a cooler can handle the default frequency as well as frequencies and voltages significantly above the default.
Intel:
- 2.4GHz @ 1.25V – Default frequency and voltage for the Q6600.
- 3.0GHz @ 1.25V – Increased frequency at default voltage.
- 3.2GHz @ 1.30V – Greatly increased frequency with a small vCore increase. Northbridge voltage increased to 1.40V as well.
- 3.4GHz @ 1.40V – Greatly increased frequency and voltage. This configuration really tests the limits of air-cooling. Not all coolers will be able to cool the Q6600 at this level. Northbridge voltage increased to 1.40V as well.
AMD:
- 3.0GHz @ 1.35V – Default frequency and voltage for the 6000+
- 3.1GHz @ 1.40V – Small increase in frequency and voltage.
- 3.3GHz @ 1.55V – Significant increase in voltage. This is a hot configuration. It is unfortunate that the 6000+ is unable to scale beyond 3.3GHz but the high voltage equates to a great deal of heat. I don’t expect that all coolers will be able to handle this load.
Noise
At this time, Icrontic will not be taking SPL (sound pressure level) measurements during HSF testing. The vast majority of enthusiast grade heatsinks have interchangeable fans, and the buyer can make the heatsink as loud or as quiet as they chose through fan selection. Our 120mm reference fan is very quiet and can barely be heard under normal conditions. The design of the heatsink can contribute to varying acoustic properties so qualitative observations will be noted in the respective reviews. If a heatsink does not have an interchangeable fan, qualitative observations will be made.
Articles RSS