Performance Results
Now for the moment you’ve been waiting for — pretty graphs! I must admit that I thought testing six heatsinks would be a quick ordeal — it wasn’t. It took me weeks to develop a reliable testing methodology and to conduct the tests. Be sure to check out the Testing Methodology section if you have any questions about what is depicted below — they will probably be answered there. It took a while to get a good hold on ambient temperature control as well — ambient temperature readings for every single test are included in the comments section (they stayed within a half degree of the mean). Please keep in mind that for each test, I have included results as reported by CoreTemp 0.95.4 as well as ASUS PC Probe/ITE Smart Guardian. The CoreTemp readings are for the hottest core (always Core 0 on the Q6600 as well as the AMD 6000+). The Smart Guardian results need to be taken with a grain of salt — they are good for comparative purposes but are far from accurate. As you will see below, the idle temperatures are reporting below ambient and clearly not correct. They are at least 7°C too low. At any rate, consistency was maintained through all the tests.
As you can see, the Thermalright SI-128 SE didn’t even break 60°C during a quad instance Prime95 load — very impressive. The Scythe Mugen and Tuniq Tower 120 trail not far behind by only a few degrees. Amazingly, the Ninja Mini and much more reasonably sized Coolink Silenator managed to stay right in the middle of the pack — even at 3.4GHz. The Kama Cross trailed behind the pack but was leaps and bounds ahead of the Retail Intel HSF. It is amazing that the Retail sink couldn’t even keep load temperatures under 60°C without any overclock at all. The Kama Cross held a lower temperature at 3.2GHz!
We see a very different picture during Small FFT Prime95 loads. It is amazing just how much more this test seems to stress the Q6600. I believe this is due to the smaller data size being able to fit into the large L2 cache of the Q6600. There is little or no bottleneck to system memory for this test. Thankfully this type of CPU load is not very representative of real-world situations. Even very intensive tasks like media encoding and folding will be accessing main memory quite a bit. The Large FFT test is more representative of what you can expect during these tasks. It should be noted that every heatsink except the Thermalright SI-128 SE exceeded Intel’s maximum thermal specification of 71°C during the 3.4GHz test. At 3.2GHz and lower vCore, even the Kama Cross was able to stay below this threshold. Again, the retail HSF did terribly. It was only 1°C away from the maximum thermal specification at default clock speed. Intel’s choice of HSF was clearly not suitable for quad core processors.
For our 4x F@H instance test, we see results not too far off from the Prime95 Large FFT testing. About a 2-3°C increase was seen across the board. The 3.4GHz configuration is clearly very demanding, even for the best heatsinks in the roundup. Nonetheless, at 3.2GHz and 1.30V, all of our aftermarket heatsinks are capable performers. The retail Intel HSF continues to trail the pack by 10°C.
Aside from the retail Intel HSF, idle temperatures varied very little from heatsink to heatsink. What is depicted is not a good indication of overall heatsink performance but is included for completeness sake.
I included “Mainboard” temperature readings for our Asus P5K-E as the different heatsinks influenced this temperature quite a bit. The sensor is located at the top right corner of the board as can be seen hereAsus Support. Even though I mounted the Kama Cross to provide airflow more towards the northbridge, the gap in the middle clearly allowed a lot of airflow to reach the area above the memory slots. It is a clear winner when it comes to keeping components on our P5K-E cool. As expected, the tower variety does not fair so well in this regard. The airflow is used almost exclusively for cooling the CPU. The Ninja Mini did very well but because of its width, the fan was literally right next to the sensor. I’m guessing that the 92mm fan’s slight overlap allowed some airflow to that area of the board.
AMD Performance Results
Intel’s Core 2 Duo and Quad processors are very popular these days but there are still quite a few individuals using either 754/939 or AM2 based systems. As such, to get a fair and complete view of a heatsinks performance, it must be compared on both platforms. Please look at our “Testing Methodology”
section for more detail on our system configuration.
As you can see, things look very different on the AM2 platform. I was very glad that I conducted testing on both 775 and AM2 because the results were surprising. The Thermalright SI-128 SE that dominated the Intel charts moved to third place, while the Mugen nudged its way right to the top. I was pleased to see the consistently good performance of the Mugen across both platforms. The biggest surprise was the Coolink Silenator. Its performance was just astounding on the AMD system. It was often only a degree off from the Mugen and is literally half its size and weight. I retested the Silenator twice on the Intel system after seeing this result but it came out the same. It is unfortunate that its AM2 performance could not be translated over to our socket 775 system. On the flip side, the Tuniq Tower 120 that did so well on our Intel system took a trip right to the back of the pack. This did not surprise me, as the mounting system for AM2 systems was nowhere near as secure as the backplate and bolt down mount for socket 775. The dislike I expressed for the AM2 clip in the installation section for the Tower 120 clearly materialized in the results. Again, retesting did not yield better results. The Kama Cross does a little better against the competition and the Ninja Mini does a little worse but overall, all three Scythe heatsinks are fairly consistent across both platforms. The retail AMD heatsink trailed the pack by about 10°C, much like the Intel HSF.
Unlike the Q6600, Small FFT loads are actually not as stressful to the X2 6000+. I speculate that this is due to less stress on the “Integrated Memory Controller.” The X2 is not bottlenecked as much as the Q6600 when accessing main memory so the Large FFT test is simply more intensive. The same trend continues with the Mugen clearly leading followed very closely by the Silenator.
Again, we see a similar trend during the F@H test.
The Smart Guardian readings are obviously incorrect. I have included them only for completeness sake and for comparative purposes. Much like our Intel system, the idle temperature varies little from heatsink to heatsink. It’s not until significant load is placed on the CPU that each heatsink can really be distinguished.
Articles RSS