With all the hardware news, it’s good to stop and look back sometimes. In cooperation with the now-defunct ExtraHardware.cz magazine, we once created the biggest cooler comparison on the Internet. However, until now it has never been published in English. So finally. A large part of the coolers are still up-to-date (including Noctua NH-D15) but there are also some historical models such as Thermaltake Sonic Tower or the first Scythe Mugen.
How time passed
In about five years of work, we have been able to compare almost all the coolers that mean or have meant something on the market with each other. Many of them are no longer sold, but the idea of their performance compared to current models is attractive.
The long charts capture the rise of SilentiumPC (Fera) coolers, but also the woes (with the older Fortis 2 and 3 clearly cannibalized by the Fera 1 and 2, and we saw some rebound in efficiency only with the Fortis 5). It is also clear from the tests why the once popular Cooler Master 212 (Plus) has fallen into oblivion or degress of Tranquillo coolers. Revision 3 was weaker than revision 2, which marked the beginning of the end of the great popularity of these coolers.
The fact that a higher heatsink weight does not automatically mean higher cooling performance is evidenced by the results of the Thermalright Silver Arrow SB-E. It’s a shame that Thermalright never followed up on this cooler, because it’s the only one that clearly beats the Noctua NH-D15 in performance, and yet the Silver Arrow SB-E is a third lighter. How so? Thermalright’s heatpipes make more efficient use of the available fin area.
In the tests you will also find the forefather of twin-tower coolers – Thermaltake Sonic Tower. However, its design is rather imperfect, with low thermal performance for current processors. The slightly newer Scythe Mugen (Infinity) could easily be sold today, although its efficiency has long been surpassed. But the Scythe series of coolers was built on good foundations from the beginning, and thanks to that the brand (Mugen) has survived to this day.
Testing methodology
The coolers are tested at maximum power (12 V/100% PWM intensity), at reduced power (fan speed is slowed down by about a third) or with the same noise level set to a low 33 dBA. The fans on the coolers are tuned to the appropriate speed in the anechoic chamber. Voltcraft SL-100 noise meter is used for measurement.
Unusual tests include performance measurements using the same Nanoxia FX12-2000 fan (at a reduced speed of 1200 rpm, but also at a maximum of 2000 rpm, and in a third mode and in combination with increased system cooling airflow). This is partly indicative of the heatsink performance. Of course, using a different fan, the order might be a bit different. Each heatsink may be better suited to a different fan. And then there are two more modes that we carry out at reduced CPU power (without overclocking) – passive mode (but with active system cooling) and
automatic regulation mode.
All tests are run on a Core i7-920 processor with a TDP of 130W. Outside of passive mode and auto-regulation, the processor is overclocked to 3.9 GHz (1.406 V) and draws around 220 W (similar to today’s Intel Core i7/9 processors without power limits).
Instead of a case, a wind tunnel is used with a plate installed vertically on the wall of the testlab. Air circulation is provided by a quartet of Noiseblocker NB-BlackSilentFan XL1 fans slowed down to approximately 550 rpm. They are spaced in a balanced ratio of two (intake) to two (exhaust). The intake is traditionally temperature controlled. This was in the range of 21 – 21.3 °C.
In the individual chapters you will find the measurements corresponding to the maximum performance of the coolers, i.e. with the speed at 100%. The quieter mode is with airflow throttled to about 2/3. The performance at the lowest noise level is measured in the fixed 33 dBA mode, when the coolers are tuned to the same noise level. To get a better idea of the heatsinks’ performance, Chapter Six shows the outputs when fitted with the same fan. These are replaced (in equal numbers) with the original ones supplied with the cooler. All these measurements are run on an overclocked Core i7-920@3.9 GHz (1.406 V) processor with power of around 220 W. But also interesting is the passive mode, in which the processor is returned to factory settings. Under such conditions, the “auto-regulation” by pulse modulation is also tested.
The load is simulated by Intel BurnTest (5 GB) for 15 minutes. If there is a result bar missing from the graphs for any cooler, it means that it did not handle that mode. Typically because it didn’t have enough performance (this applies to passive mode at least) or it had an atypical build and it was not possible to fit a standard 120mm fan on it.
English translation and edit by Jozef Dudáš
I see numerous inconsistencies in the results. I don’t think the test has an accuracy of one tenth of a degree Celsius. Based on the example below, I would estimate an accuracy of 2-3C. However, since the coolers are often only 0.1C apart,
the test methodology obviously does not allow for such accuracy, possibly only 2-3C as can be seen in the following example. I fear that, at best, the results of this test are only suitable for obtaining a rough overview of whether a cooler is to be classified in the upper, middle, or lower segment in terms of performance and noise level.
For a scientifically accurate ranking, several tests would have had to be carried out for each sample, in which the test setup would have to be repeated (repeat assembly and reapply paste). The results would probably only be useful if several samples of each model were tested.
However, despite such a financial and time investment, it is doubtful whether your equipment and test setup would even be capable of producing reproducible results with such accuracy.
Then the result would certainly not be that, at a load of approx. 220W, an NH-D15S performs 2 degrees Celsius better in the burn-in test than an NH-D15, even though they are identical in construction and have the same fan. The only difference is that the NH-D15 model has 2 fans instead of one.
We never claimed that the accuracy of results is at the 0.1 °C level—that’s something you’re bringing up. And nobody can achieve that in a natural environment anyway, simply because of the measurement error of the CPU’s internal temperature sensor. Anyone who claims otherwise either doesn’t know or is just making things up.
I disagree that the measurements we did in the past were insufficiently accurate. It’s been some time, really many years ago, but I remember that repeated measurements scaled excellently. That was something we were really strict about. I know this quite precisely. 🙂
„For a scientifically accurate ranking, several tests would have had to be carried out for each sample, in which the test setup would have to be repeated (repeat assembly and reapply paste).“
Yes, exactly—hat’s what we did. Multiple samples, repeated assemblies. You know this process well. 🙂
„Then the result would certainly not be that, at a load of approx. 220W, an NH-D15S performs 2 degrees Celsius better in the burn-in test than an NH-D15…“
You certainly measured that correctly. 😉
I think he was referring to the test with automatic regulation,
https://www.hwcooling.net/en/gigantic-retrotest-of-80-cpu-coolers-2005-2015/6/
…where, among other things, the D15S has a temperature of 57°C in burn-test and a noise level of 36 dBA
and the D15 has a temperature of 62°C and a noise level of 37.3 dBA
in idle: D15S – 32°C (32.8 dBA) D15 – 35.5°C (33.4 dBA)
… this seems like an error in the data to me too 😛
…
while in other tests the D15 dominates over the D15S
by other tests I mean: maximum speed, reduced speed and 33 dBA
Now, more than 10 years after the tests were done, I have no way to verify anything anymore. In any case, the “automatic regulation” measurements are at lower load levels of a non-overclocked CPU. That also applies to the passive mode scenario. The other results correspond to the higher power draw of an overclocked CPU, just as (hopefully) stated in the testing methodology. It’s true that within this particular article the methodology description is rather poor. Certainly poorer than it used to be in the original tests on ExtraHardware.cz. These, here on HWC, are mainly an archive of something that would have been a shame to lose. 🙂
“We never claimed that the accuracy of results is at the 0.1 °C level—that’s something you’re bringing up.”
I have reread your paragraph on test mythology. The topic of accuracy was not addressed at all. Therefore, the accuracy you claim can only be determined based on the test results. (These are fundamental and generally accepted principles.) Here, you claim an accuracy of one tenth of a degree Celsius. Perhaps you should round the results to a full degree Celsius and rearrange the tables accordingly in order to present results that are at least approximately comprehensible. However, I doubt that this would be sufficient, since, as I said, the accuracy of your measurement method is at best approximately 3 degrees Celsius.
“Yes, exactly—hat’s what we did. Multiple samples, repeated assemblies. You know this process well. 🙂”
This is an important step in producing scientifically sound results. But if you have gone to such extra effort, in terms of time alone, this can easily mean twice the work. Then surely you would have advertised this in the foreword to the results? Personally, I would be extremely proud of this and would report exactly how many tests I had carried out per specimen and whether the value in the finished table is an average value between all of them or whether only the highest and lowest values were counted, or whether you calculated all values except the lowest and highest.
Unfortunately, I have not read your print editions due to the language barrier, but I am sure that such details were explained there?
Why am i telling you this? The average Joe that is looking for the right cooler for his Retro Build will go to your Graphs and see WOW this cooler performs 0.4 Degrees better than this Cooler. And base his buying decision on that. But if the displayed accuracy is so good that actually “nobody can achieve that in a natural environment anyway” the reader gets bamboozled.
Just to be sure, I don’t know if you noticed, but these tests were conducted between 2005 and 2015 for the now defunct portal ehw.cz. 😉 so…
This test only represents an archive of results from tests that were originally published in a completely different magazine. I am the author, and I explain why in this case there isn’t such a detailed testing methodology as HWCooling readers are used to. It’s great that you are demanding.
Yes, you’re right. The accuracy of the CPU’s internal sensor is too low to talk about “super precise” results. However, I don’t think readers were necessarily misled. The pursuit of extreme precision is not what defines the conclusions of tests. Even though the CPU sensor is less accurate compared to artificial test environments, I still think these results are more useful to the end user. Because they reflect a more realistic situation. Of course, the best approach would be to test all coolers with all CPUs, at different load intensities. But in practice, that’s all but impossible to execute. 🙂
Who knows how would the TR SA IB-E Extreme rev.B perform in the tests. I’m guessing it would do better on an Intel platform than on a small AMD, especially with its large base. 🙂
That’s not something we can easily find out. The cooler does support the AM5 platform, but for Intel it only goes up to LGA 1200 (and LGA 1851, resp. LGA 1700 are missing, both with different mounting hole spacing). The performance ratios will naturally vary from processor to processor, since the heat distribution from CPU to cooler differs. And so the requirements for contact quality and the overall base design with heatpipe layout differ as well. Still, I’d also be very interested to see where those coolers would stand today—not just the SA IB-E Extreme rev. B, but also the SA SB-E (Extreme). 🙂
They have in offer – at least on their site – plenty of LGA 1700 mounting kits and one supporting SA IB-E Extreme rev.B is LGA1700-B2. Four years ago they posted that man can get kits by contacting their support. Not sure if links are allowed here, so just copy to search bar “THERMALRIGHT upgrade mounting kit supports Intel LGA1700 CPU”. I wonder how it works, so if it still works and how much it is. In theory they support many of their discontinued coolers what is great and enables amazing follow up to this article – comparing on modern platform some oldest or legendary coolers like Silver Arrows and D14 to latest and greatest Thermalright’s and Noctua’s offerings to show us how much air cooling moved forward over the years :> Especially given that at least to my knowledge nobody has ever done something like that.
You can give a link 😉 (but if there are more than one, they have to be approved)
This is for SA SB-E/IB-E
https://www.thermalright.com/product/lga1700-b2/
…and there is also one for rev.B 😉
So you can split them into two comments. 😉
https://www.thermalright.com/product/lga1700-b2-2/
Thank you! 🙂
Thanks for the info. The only thing left is to get our hands on SA coolers and find some time to test them. 🙂
Sorry, I think there has been a misunderstanding. So, when you talk about a magazine, you mean a website, not print media? And every individual cooler tested just once and not multiple times? Right? As I understand it, all the tests were conducted with the same equipment (microphone, chamber, mainboard and CPU), but over the course of many years.
Do you have any Explanation why the NH-D15S sometimes performs 5°C worse than the NH-D15 (220w burn in) and sometimes performs 5°C better (auto regulation burn in)? i mean, Theoretically, that could be a difference of 10°C with the same cooler. Was it just a typo?
I know Noctua 140 mm coolers very well, and yes, the NH-U14S can outperform the NH-D15 in terms of performance under noise-normalised tests. But there’s no way the NH-D15S can outperform the NH-D15 with the same fan and rpm.
PS: In my experience, the processor temperatures shown in e.g HWinfo are extremely useful. I can immediately reproduce temperatures of less than half a degree Celsius by switching between different thermal pastes or installing different pumps in the circuit. And this has been working for over 10 generations.
Yes, when we say “magazine,” we actually mean “website” (ExtraHardware.cz was that site…).
Every cooler was tested multiple times, or at least we know that the repeatability of the measurements had a very small margin of error. Small enough that it did not distort or change the conclusions. I made sure of that even back then. In the past, we also studied not only how cooling performance is affected by different amounts of thermal paste applied with different techniques, but also what impact mounting pressure has. It’s not a groundbreaking test, but it does provide some orientation:
https://www.hwcooling.net/en/test-how-can-a-pressure-of-a-cooler-affect-its-cooling-performance/
— „Do you have any Explanation why the NH-D15S sometimes performs 5°C worse than the NH-D15 (220w burn in) and sometimes performs 5°C better (auto regulation burn in)?“
When exactly does this happen? Please be specific, so we can refer to a particular situation our tests represent. If in one case the Noctua NH-D15 leads and in another the NH-D15S, then perhaps it’s due to different thermal loads? As mentioned in the earlier comment, some scenarios were tested with lower power consumption of a non-overclocked CPU.
— „But there’s no way the NH-D15S can outperform the NH-D15 with the same fan and rpm.“
At the same noise level, the achieved fan speeds will not be comparable. A cooler with a single fan will have its fan running faster. What you’re saying can of course be true, but you can’t compare two fans against one. The NH-D15 (with two fans) naturally has the potential for higher cooling efficiency, or for achieving the same cooling performance at lower noise. Its dual-fan design predestines it for such results – though that may not always be the case. From experience we know that the additional fan at the edge doesn’t always significantly improve cooling, and at equal noise levels it could, under certain conditions (specific load on a specific CPU), even turn out that the NH-D15S comes out ahead.
— „I know Noctua 140 mm coolers very well…“
We would really appreciate it if you could share your extensive experience in the form of exact measurements and specific figures under specific conditions. 🙂
Two People have told you exactly in which Graphs you can find this numbers at 3 occasions. As long as i am not able to upload Screenshots in the Comment section i am afraid you will have to look at the Graphs yourself. And no we do know how to read Graphs and none of the tests where noise normalized.
“Bufo11. 9. 2025 at 14:58
I think he was referring to the test with automatic regulation,
https://www.hwcooling.net/en/gigantic-retrotest-of-80-cpu-coolers-2005-2015/6/
…where, among other things, the D15S has a temperature of 57°C in burn-test and a noise level of 36 dBA
and the D15 has a temperature of 62°C and a noise level of 37.3 dBA
in idle: D15S – 32°C (32.8 dBA) D15 – 35.5°C (33.4 dBA)
… this seems like an error in the data to me too 😛
…
while in other tests the D15 dominates over the D15S”
“Jesus Official15. 9. 2025 at 22:39 …Do you have any Explanation why the NH-D15S sometimes performs 5°C worse than the NH-D15 (220w burn in) and sometimes performs 5°C better (auto regulation burn in)? i mean, Theoretically, that could be a difference of 10°C with the same cooler. Was it just a typo? …”
“We would really appreciate it if you could share your extensive experience in the form of exact measurements and specific figures under specific conditions.”
As soon as i notice a test that has not already been performed and documented somewhere in a Blog, Forum, Google Spreadsheet, Website or YouTube Video and that can easily be found with a Search engine i will think about it.
Thank you for your response. Of course, we can look at this from different angles. Was it a mistake, was it not? As I already mentioned, we can no longer determine that now. These tests were carried out roughly 10 years ago, and the accuracy of the results can no longer be verified (typically by re-testing the specific measurement).
I certainly would never claim any kind of dominance of the NH-D15S over the NH-D15, although yes, in the automatic/PWM regulation tests at lower power consumption, the single-fan cooler is shown with somewhat more attractive results (lower temperatures at lower noise). Can you factually disprove that this cannot happen? Specifically in that PWM regulation mode, where, at lower CPU power, only part of the NH-D15’s potential is being used compared to the NH-D15S. Personally, even after all my experience with cooler testing, I would not dare to state outright that the presented situation could never occur in practice—but perhaps you see it differently? If you can provide a factual explanation on scientific grounds as to why this result definitely cannot happen, that would be excellent.
Regarding the last paragraph of your post: please show us what you have available, so that we have something concrete to work from. Regardless of whether someone else has already published something similar, we would very much like to see your work, which clearly confirms what you claim in your text. Lay your cards on the table as well. 🙂
Well, I actually expected you to form your own opinion and look at the graphs after we had pointed out the obvious measurement errors to you several times. Which you didn’t do. So I can hardly expect you to check again whether these erroneous results were already present in the original publication in the magazine or whether it might be a transcription error. Because I would assume that these results were not published that way at the time. Anyone would have noticed immediately that it is not possible for the NH-D15S to perform several degrees Celsius better than the NH-D15 with the same fan and the same RPM when idle. And please keep in mind that the Intel burn test was also measured in the PWM test. What was the TDP again?
What information should I share with you here? I currently have no documentation of my results, apart from a few HWInfo screenshots from 10 years ago. Why not, you might ask? Because several content creators have already produced exactly the same results. The NH-U14S can be quieter than the NH-D15S under low heat loads, e.g., when doing office work or browsing, or when running with a low TDP Intel T processor. That’s something to remember 😁👍 . This can be explained by the fact that half as many fins have to be passed through by the air. The dual tower coolers only shine at higher TDPs.
It is not a mistranslation, and the charts were published exactly like this at the time. Here is a version of the article in the successor portal cnews.cz (of ehw.cz).
https://www.cnews.cz/clanky/gigatest-100-chladicov-ktory-kedy-a-v-com-ma-navrch-ci-taha-za-kratsi-koniec/
The problem is that Ľubo probably no longer has the logs from that time to check it. So he can’t say that it is a mistake or that it is not a mistake (he can guess, but he doesn’t like to do it publicly). 😀
But let’s not get into this old article, which is here just for the memories 😉
I still have the records from that time, I went through them (after those paper magazines Hardware Special, yes) and on the level of presentation everything is certainly in order. There were no incorrect transcriptions… but I can’t re-test it anymore, so we won’t be able to verify it now. Who knows what has happened to that processor (the Core i7-920) today—I don’t even have it anymore, and haven’t for quite a long time. 🙂
Of course, anyone can dive into whatever they want, but when it comes to fabrications and assumptions… well, everyone works with what they have at hand. I believe that real value comes only from information one knows. To “think” is something anyone can do. It’s nice, but as Max Kašparů once said (and in my opinion he put it beautifully): “A layman may think, but an expert must know.” And I agree with that. 🙂
Thank you for the response. However, I must point out that you are ignoring the points made in the comments and instead creating your own narratives. Given your inability to provide any sort of evidence, your statements about erroneous results are quite bold to say the least. This sort of manipulation might work elsewhere, but I’m afraid it won’t pass in a discussion on HWCooling.net. 🙂
What would be helpful is if you could actually provide some factual evidence as to why the results must be wrong. Since it seems you are the knowledgeable one here, I’m sure you’ll able to clarify. Please do so scientifically, factually. Feelings are fine, but what matters here are facts.
By the way, it looks like Nanoxia is dead from the website, 🙁
https://www.nanoxia-world.com/page/home.php
although you can still buy boxes here and there…
What do you know about it, Ľubo?
I don’t know anything about the (non)existence of Nanoxia, but I did notice recently that their website stopped working. I still have some FX-2000 fans lying around somewhere. 😀
I have addressed each point in detail. What would constitute sufficient evidence for you? Should I begin by linking the first 100 tests appearing on Google comparing the NH-D15 and the NH-D15S in the comments? Because no one has results showing that the NH-D15S performs 5C better in the IDLE than the NH-D15. Such results can only be found on hwcooling.net.
EDIT: 3,5°C
Thanks. If you’re unable to technically substantiate your claim about an error, then I’ll share my own theory as to why this situation (with the NH-D15S showing higher cooling performance at lower power draw of the i7-920) can indeed occur in practice. 🙂
The fact that we are the only ones who measured this (on HWCooling.net) does not automatically mean it’s incorrect. Yes, precise cooler testing is on the decline, with various marketing videos and similar materials taking its place—but those wouldn’t uncover such behavior anyway.
Please break it down on a scientific level (technical reasons…) why the results presented in the tests could not occur in the given situation. You certainly haven’t done this in any of your comments so far. If I’m mistaken and I’ve overlooked something, then please point me to that specific comment. 🙂
Let’s get into it: Why can the NH-D15S (with a single fan) outperform the dual-fan NH-D15 under lower load conditions — while at higher load (when more heat needs to be dissipated from the CPU) the situation can be different?
Here’s the “scientific” explanation I mentioned before, which you haven’t provided. I’ll try to keep it simple and accessible.
The efficiency of a cooler’s heatsink can vary depending on the thermal load. Its performance characteristics are different at low and high loads. The reason is that with varying thermal output, the heatsink’s efficiency (as a combined system of heatsink + fan) scales differently. That means that under lower TDP conditions, the NH-D15S can, under certain circumstances, have the upper hand—while at higher loads, it may fall behind the NH-D15 in cooling efficiency. This can certainly happen, since the heatsink operates differently at different thermal intensities. At low CPU load, the front fan and the cooling of the front tower might not even contribute much to the total thermal transfer. It can happen—or not—depending on the specific CPU model used. It’s individual.
So, I definitely wouldn’t call the presented results an “error” unless you’re providing opposing conclusions measured under identical conditions. 🙂