Our Noctua NH-U12A cools better with one fan than with two – but why?!

The new Noctua NH-U12A was also in our test system. However, we found some discrepancies in extended tests. It’s troubleshooting time.

Our Noctua NH-U12A test – or what’s left of it

With the NH-U12A, the Austrian company Noctua has launched a new premium cooler on the market. With two NF-A12x25 fans, a wider heat sink and an absolute premium price, however, it is only the figurative successor of the NH-U12S, which plays in a completely different class with a narrow heat sink and only one fan. The NH-U12A, on the other hand, can even compete with twin tower coolers thanks to its new design. However, the cooler does not by far need its space and is designed for maximum compatibility with PCIe and RAM slots. The highlights are the two high performance fans NF-A12x25, which have been developed by Noctua for highest performance for many years and already have an impressive price tag.

The combination was of course also interesting for our tests. Accordingly, we obtained a test sample from Noctua and installed it in our test system. The test of the NH-U12A turned out a little different than expected. In comparison with its brothers NH-U12S, NH-U14S and NH-D15, it is subject to the two 140 mm variants. The new radiator is able to assert itself against the predecessor NH-U12S, but also here the distance is not as good as expected. The tests from other magazines also spoke against our results. Noctua promptly sent us another NH-U12A to rule out a bad model or an error. We took the chance and started our test all over again and made it more detailed. For the test itself we also received a lot of feedback and requests for extended tests – for example with only one NF-A12x25 fan on the heat sink or different speeds. And that’s when the inconsistencies started.

The new test

Initial situation and initial changes

In preparation for upcoming cooler tests, we have once again revised our test concept. The test series with the Noctua coolers was quite suitable for this, as we were looking for faults anyway. In our initial test, we measured with some special parameters that make the test result objective, but not always realistic. For heat generation, we used our Intel Core i7-7700K with the Small FFT test of Prime95. The Case, a Dark Base 700 from be quiet!, stood upright with closed side panels. For ventilation, we have mounted two 140 mm fans in the front and one 120 mm fan on the rear and set them to 40 percent constantly. The fans are all Silent Wings 3 models from be quiet!. Originally, we tested with a very strong fan curve, which made it much more difficult to compare the individual coolers. Therefore, in the new test series, we performed one test per cooler at 100, 75 and 50 percent speed.

Heureka, the new NH-U12A works better – but the old one too?!

In good faith now to bring a better comparable test, we have again mounted all five test coolers on our mainboard. We tested a total of seven different constellations:

• Noctua NH-U12S
• Noctua NH-U12S with NF-A12x25
• Noctua NH-U12A (first sample)
• Noctua NH-U12A (second sample)
• Noctua NH-U12A with one fan
• Noctua NH-U14S
• Noctua NH-D15

Once again, interesting results were achieved. The second test sample of the NH-U12A performed significantly better than the original sample. Suddenly it even beat the NH-U14S and NH-D15. It was noticeable, however, that the old sample of the NH-U12A performed much better in this test series too and came close to the two 140 mm coolers. Curiously, the new NH-U12A performed even better with only one fan than with two. Another interesting result was the NH-U12S. If this was equipped with the NF-A12x25 fan, it achieved a better result than the first NH-U12A at 75 and 100 percent speed in normal mode. Overclocked, the results then gapped clearly apart again, the NH-U12S with the NF-A12x25 fan was even defeated by the original fan.

Further adaptations – new system, new test series

So everything indicated that we had a big problem somewhere in our test system. Therefore, we continued to experiment to find a better test variant. The first mistakes were quickly found. So it definitely made a difference whether the processor had a cooling phase between tests or not. In our original tests we only changed the rpm and restarted Prime95, there was no cooling phase. In the new test procedure we then tested the following procedure intensively with the NH-D15:

• Allow to cool down at 100 percent speed for 10 minutes (also at system restart)
• Launch Prime95
• Warm up for one minute
• Start measurement with AIDA64
• Run 10 minutes
• Read values
• Stop Prime95
• Start of cooling phase for next test

And lo and behold – the test results only varied by 0.5 degrees Celsius. In previous tests without a cooling phase, the temperature still varied by up to 2 degrees Celsius. However, a longer cooling phase did not bring a better advantage, which is why we stayed at 10 minutes. Further we have lowered the variance then thanks to note of a user. He claimed that the Prime95 test can vary slightly even in Small FFT mode. We then started a series of tests with the new method. While the deviation in tests with the predefined Small FFT mode was still 0.5 degrees Celsius on average, a test with defined 12K in-place FFTs resulted in a deviation of only 0.2 degrees Celsius. This is how we defined our general test method. To be on the safe side, from now on we also no longer operated the fans with the mainboard, but with an external fan controller.

Next test – abort in the middle, but the faulty part is found

After that we started the test series again. The time required for the new method had now also increased considerably. For each cooler, 126 minutes of testing were now required, based purely on the test times. However, we accepted this for precise tests, above all because the tests work . But we stopped this new test series in the middle of it, because we found another variable, which was probably also responsible for the curious results of the second test.

Our test processor has been delidded and treated with liquid metal for quite some time. During this process, the heatspreader is removed, the heat conducting paste is exchanged for liquid metal and the heatspreader is glued on again, resulting in significantly better temperatures. But for our cooler tests, that was exactly the problem. The tilting of the case already seems to have caused a change in the liquid metal, which actually affected the temperature. In several tests we have measured differences of up to 3 degrees Celsius.

Okay, so started again with a new test setting. Now it’s about moving the processor as little as possible. But even with a permanently positioned case, there were always different test results when the cooler was remounted. This makes a comparison almost impossible, which is why we finally aborted the test.

The Noctua NH-U12A still cools better with a single fan

All these tests have shown us that testing coolers is anything but easy, especially with a delidded processor. One thing, however, seemed particularly strange to us. The Noctua NH-U12A cooled better with one fan than with two in all the tests we did. Only in overclocked mode at very high temperatures the constellation with two fans was superior.

We do not yet understand why this is the case. In any case, the deviation due to the delidded i7-7700K can be excluded as the cooler has not been dismounted and the case has not been moved. We only removed the rear fan. And every time there were slightly better temperatures with a single fan. For accuracy, we conducted the test several times with both NH-U12A coolers, but always came to the same result: with loads up to about 73 degrees Celsius, the NH-U12A performed better with a single fan. Noctua was also very surprised about the result. The manufacturer developed the radiator for years and probably tested it even more extensively than we did. Other tests also speak against our test, but after this extensive troubleshooting we can almost completely rule out that it is a gross blunder on our part.

Conclusion: we need a new test system for CPU coolers

The whole intensive test series around the Noctua NH-U12A brought us at least one insight. For future cooler tests, we definitely need a different test system that finally eliminates the deviation from the liquid metal. Therefore we are currently planning a Ryzen system with soldered heatspreader. The more than 140 test runs performed were not quite for nothing because of a second thing. The intensive series of tests has also enabled us to significantly improve our testing methodology.

But why the NH-U12A performs better with a single fan in some situations hasn’t been shown by the many test runs. As soon as our new test system is ready, we will start a new test series. Then we also know if the strange results were caused by our delidded i7-7700K.

Do you have a Noctua NH-U12A? Then we would be happy if you also tested Prime95 with only one fan and share the results with us in the comments here!