D15 is elite, but there’s a room for improvement

Ľubomír Samák

6 years ago

How we were testing

The ideal CPU cooler should not be bigger than a matchbox, and it should handle 500 W processor without making a sound. However, this is a sci-fi, so we will have to rely on conventional solutions. Although Noctua NH-D15 is in many ways unusual, it is remarkable to see how its original elements affect the effectiveness.

Bonus tests of coolers are a regular extension of standard reviews. In addition to cooling efficiency around the socket, they include testing and comparison of heatsinks using the same fans, as well as fully passive operation without any active cooling. And also other various tests that might be interesting for a particular cooler.

How we were testing

During the tests, we also measured CPU MOSFETs heating at 33 and 39 dBA. We put the sensor right into the heatsink. These tests always begin with the same starting temperature (35 °C). Although Gigabyte is known for an excellent cooling of its power supply and manages to maintain very attractive temperatures, it is still easy to determine which of the coolers more or less keeps doing a good job cooling the critical components.

The fans were aligned in such a way that they do not exceed over the top rib because it would be pointless there. It is more important to swirl the air in the lower parts. This was still measured with the default fans.

In the next phase of the tests, the original fans were replaced by the reference fans to create the performance comparison of the heatsinks with the same fans. These results might come in handy when you want to replace the default fans. Usually because they are too fast or too slow (and inefficient), or they produce disturbing sounds, or for whatever reason based on personal preferences.

Optimal candidates for such tests are the “industrial” Noctua NF-F12 iPPC fans. Thanks to their wide range, we could test with a very high flow and at a very low noise level. We increased the flow of the system fans (4× 1200 rpm) only in 24 V mode (2000 rpm without any problems). For a better comparison, we also included the tests with standard 4× 550 rpm. After reducing the voltage to 15 V, the fans were relatively quiet (1290 rpm) and at 9 volts they were running at 765 rpm only, which could be especially challenging for large two-tower heatsinks and larger liquid coolers. We replaced the reference fans with the number corresponding to the original set. Usually one or two. For a better orientation, it is marked in the charts (1/2 ×).

Finally, the coolers competed with each other without a fan. The processor we used was Core i7-5930K (TDP 140 W) with Vcore 0.9 V, which had 25 W lower power consumption than with default settings. These tests took longer than those before, 900 seconds. Most coolers cannot handle this. After 95 ºC, tests were interrupted and considered unsuccessful. Therefore, temperatures in idle are also interesting. We used traditional configuration of system cooling and airflow.

Extra test: This time, we will try to find out how removing of one fan affects the cooling performance of a cooler that has a wider space between its towers. Standard configuration of the cheaper NH-D15S comes with one fan only. In Chapter 4, this is also illustrated by thermal images.

The ideal CPU cooler should not be bigger than a matchbox, and it should handle 500 W processor without making a sound. However, this is a sci-fi and we will have to rely on conventional solutions. Although Noctua NH-D15 is in many ways unusual, it is remarkable to see how its original elements affect the effectiveness.

Cooling the MOSFETs


Cooler		Speed and voltage of the fan
		39 dBA	31 dBA	33 dBA	31 dBA
Arctic Freezer 33	Arctic Freezer 33	1× 1148 rpm	8.8 V	1× 848 rpm	6.23 V
SilentiumPC Fera 3 v2	SilentiumPC Fera 3 v2	1× 1393 rpm	9.4 V	1× 971 rpm	6.59 V
Noctua NH-D15		2× 1073 rpm	8.1 V	2× 700 rpm	5.87 V

/* Here you can add custom CSS for the current table */ /* Lean more about CSS: https://en.wikipedia.org/wiki/Cascading_Style_Sheets */ /* To prevent the use of styles to other tables use "#supsystic-table-91" as a base selector for example: #supsystic-table-91 { ... } #supsystic-table-91 tbody { ... } #supsystic-table-91 tbody tr { ... } */

With the same fans

Thermal images of the heatsink

Heatmaps were created before the end of each burn test while using the same fans (NF-F12 iPPC) at different RPM and in different numbers. Measurements were done in our wind tunnel as usually. We subtracted the temperatures from the part of the heatsink that is in one axis with heatpipes. Because of the problems with the reflective nickel (even after the emissivity adjustment according to the table values, the temperature reading would not be accurate), there was a strip of black insulating tape on it.

▽ 2× NF-F12 iPPC. On the left 24 V (~ 2000 rpm) and on the right 9 V (~ 765 rpm) ▽

▽ 1× NF-F12 iPPC. On the left 24 V (~ 2000 rpm) and on the right 9 V (~ 765 rpm) ▽

Note the increase of heat from the heatsink after removing the first fan. And especially with low RPM. At a high flow rate, the difference was smaller and the effectiveness of very intensive suction is thus seen. This also indicates a smaller contrast (and therefore higher heating) of the fan frame.

Passive mode (+ prints of the compound)

Prints after the disassembly

Conclusion

Let’s take it step by step. Fans exceed the heatsink quite a bit at the bottom, which obviously helps to cool VRM. But less than it may seem. The results are negligibly better compared to single-fan coolers with less overlapping.

More interesting is the analysis of tests with reference fans. A good comparison can be made at 9 V (41.9 dBA) when the cooling performance at the same noise level is lower (by 2.3 °C) than with original two NF-A15. Replacing the fans by smaller ones because you want to create a space for higher RAM (+ 2 cm) does not pay off. With further noise reduction, this difference is, of course, even more visible. And it becomes significant if you want to remove the side fan completely. Other dual tower coolers do not suffer that much as D15 when you remove one fan, but here it can make + 5 °C if you want to have a really silent cooling.

Due to the centimeter gap between the rotor and the tower, the suction effect is very weak, which is confirmed by the thermal images. But everything bad is good for something, and thanks to this slit, the noise of the cooler at the same flow is definitely lower than if the fan was attached right to the ribs. The second tower simply creates less flow resistance. The proof that it works is the comparison of the measured noise with coolers that have only one tower. Almost the same results were measured with Arctic Freezer 33. Sure, D15 has a whole range of noise reduction elements and tests showed that they do improve things a bit indeed.

However, the mentioned above is more or less trivial. Cooling efficiency (in terms of power/noise) is the domain of Noctua, no doubt about that. It is the heatsink that is not that excellent (in terms of effectiveness/weight ratio). That is not 100% supported by our tests yet, but we will prove and explain everything once we get our hands on Thermalright Silver Arrow SB-E again. This cooler, by the way, has the most effective heatsink we have ever seen (during some previous work). It’s 250g lighter and has a very similar efficiency as D15. Unfortunately, TR has buried the series some time ago.

We have to mention high performance in the passive mode. In conventional cases that have the upper air outlet out of the axis of the ribs, it can be even better than Zalman FX70. However, in general, all dual tower coolers are suitable for passive cooling.

If you would like to give us a tip for a review, feel free to do so in the comment section.
We will be grateful for any feedback!

Continue: Cooling the MOSFETs