We are overclockers. We are not satisfied with stock. Like our automotive forebears we hotrod our rigs, pushing them to faster speeds. That’s what we do. And when we do it, our chips get hot and we need to cool them off.
Let’s say you really want to overclock your CPU, and you want to cool it with air. You know that the Noctua NH-D14 is a very good heatsink, designed for quiet cooling; but you want more. Can you get better cooling from a D14? Let us find out.
About the NH-D14
First, you may ask what is a NH-D14? A Noctua NH-D14 is a device that keeps your CPU from overheating, even when it is overclocked. The D14 is made up of a nickel-plated copper contact plate that is soldered to six heatpipes. Each heatpipe leaves the contact plate in two directions, bending away from the motherboard and piercing two stacks of fins, one with each end of the heatpipe. So you have two stacks of fins about an inch apart, connected at the bottom where their heatpipes pass through the contact plate. It sounds fragile, but this is a very sturdy device. Without fans, the D14 has no moving parts, though there is vapor inside the heatpipes.
When the contact plate is bolted onto the integrated heat spreader (IHS) of a CPU that has a bit of thermal interface material (TIM) on it, the heat from the CPU is transmitted to the heatpipes, which in turn transmit it to the fins, which radiate the heat to the air. Now, you could leave it that way and the heated air would move by convection, allowing the fins to transfer their heat to the air that wafts in. Or, you could put the system in a case with air flowing through it, which would allow the fins to shed even more heat. This is known as running the heatsink as a passive cooler; the D14 works pretty well this way.
If you want to get better cooling you will do what Noctua did: put fans on it. Noctua chose to have a 120 x 25 mm “push” fan clipped to the first finstack, blowing air through the finstack toward the center gap and the center fan. In the center gap there is a 140 x 25 mm fan that is clipped to the second finstack. The center fan pulls the air coming from the first stack and pushes it into the second stack. The stock system cools a CPU very well and does so quietly. Of course, overclockers always want more. Noctua understands this, and provides a third set of clips and pads if you write to them. But what if we want more powerful fans?
I tested a number of fans and combinations on an overclocked system cooled by a D14. I recorded the noise that the fans made as they were blowing air through the D14. I recorded the cooling performance. Below you can look at the results and decide for yourself whether upgrading the fans on a D14 is worth doing. The results will apply not only to a D14, but also to any other tandem-finstack cooler.
|CPU||Intel i7 860 @ 4004 MHz with 1.312-1.328 V vcore|
|Motherboard||GA-P55A-UD3P (sitting in open air)|
|RAM||G.Skill Ripjaws DDR3-2000 @ 1456 MHz|
|Graphics Card||PowerColor AX3450 Radeon HD 3450 (fanless, hence noiseless)|
|Solid State Drive||Kingston SSDNow V+100 64GB|
|Power Supply||SeaSonic X750 (The fan mostly doesn’t run with the above load.)|
|Heatsink||Noctua NH-D14, with Gelid GC Extreme TIM|
|Stress Software||OCCT 3.10|
|Tenma 72-942 Sound Pressure Level Meter|
|Digital TEMPer USB Thermometer|
By now, the NH-D14 has been in place for more than a year so that temps can be compared between fan runs separated by months. When identical setups are periodically tested, the results are very close to each other.
The TEMPer USB Thermometer is set 30 cm in front of the intake finstack. The thermometer software records a log that is saved as a spreadsheet. A mean ambient temperature during each run is calculated and recorded to the nearest tenth of a degree centigrade.
OCCT acts as a front end for Intel’s Linpack, which was written by their engineers to maximally stress their chips. OCCT also records a log that can be saved as a spreadsheet. Since Linpack tests the CPU in bursts, the hottest core temps are harvested and a mean CPU temperature is calculated, and reported to the nearest tenth of a degree centigrade. Temperature Over Ambient (TOA) is calculated by subtracting the mean Ambient temp from the mean Core0 temp (Core0 is the hottest core for this chip).
The Tenma SPL meter can measure SPL in the A and C scale. I recorded measurements in the A scale, notated as dBA. The SPL was measured at 10cm from the side of the center fan (or where the center fan would be). This allowed for a measurement position that was consistent regardless of fan setup. The 10 cm SPL’s were then correlated with 1 m SPL’s by subtracting 20 dB. The results are reported as if the sound levels were measured from 1 m. This allows you to compare sound levels with those in other reviews. Results are reported to the nearest half dB.
My rule of thumb scale for noise:
|Relative Sound||Sound Pressure Level|
|Too Loud||greater than 60 dB|
|Very Loud||50-60 dB|
|Moderately Loud||35-40 dB|
|Very Quiet||18-24 dB|
|Ultra-Quiet||18 dB or less|
There are a variety of combinations of 120 x 25 mm and 140 x 25 mm fans that can produce noise that ranges from Ultra-Quiet to Moderate. We will be looking at combinations of fans that will make the system range from Moderate to Very Loud: in this set of tests we are more interested in excellent cooling than quiet operation.
The Fans: The stock D14 comes with a 120 x 25 mm push fan and a 140 x 25 mm center fan. The center gap won’t accommodate a fan thicker than 25 mm, but that doesn’t keep us from putting 38 mm fans on the front (push) and back (pull) of a D14, with or without a center 25 mm fan.
The cooling results of various fans are reported separately by blade count and by thickness. The 25 mm fan cooling results are analyzed separately for push+center and push+center+pull setups. The 38 mm fan setups are each considered as groups, to examine the effects of push+center vs. push+center+pull vs. other setups.
The Charts: The charts below will name the fans in order, from push to center to pull, state the rpm of each fan, then show the SPL (in blue) and the TOA (in peach). In the first chart, the number shown in parentheses is the number of blades of the center fan.
The Prelims – Solo Center Fans and the Stock Setup
I would like to introduce you to the fans that will sit in the center of the heatsink, between the two finstacks. The D14 comes with a 140 x 25 mm fan in the middle that has screw holes set for 120 mm fans. This means that any fan that is 25 mm thick and has 120 mm screw holes will fit. So of course you can use a 120 x 25 mm fan in the middle. However, all fans that you want to use in the center position of the D14 must be modded. The modification is fairly easy. I’ve gotten so I modify a fan in about a minute, but you must take care.
Stock fans: Noctua ships the D14 with a nine-bladed 120 x 25 mm NF-P12 push fan rated at 1300 RPM, and a 140 x 25 mm NF-P14 center fan rated at 1200 RPM. Both fans have “SSO bearings.” To understand what that means you must go to the Noctua.at website. I can’t do it justice.
Kaze Maru 2 – 1700: A nine-bladed 140 mm fan from Scythe, rated at 1700 RPM. It’s model SM1425SL12H, has a sleeve bearing and screw holes that match those for 120 mm fans. So it is perfectly compatible with the D14 — after you do a little modding (link).
Scythe Slip Stream – 1900: Scythe Model SY1225SL12SH. It is a nine-bladed 120 x 25 mm sleeve bearing fan rated at 1900 RPM.
Thermaltake A2018: It is a seven-bladed 120 x 25 mm fan rated at 2800 RPM. The box says it has one ball bearing and one sleeve. It may look like a sped-up Yate Loon fan, but it is not. It has pretty blue LED’s and will draw blood if you let it bite you.
Gentle Typhoon AP-30: Gentle Typhoon “for high speed applications,” a seven-bladed 120 x 25 mm ball bearing fan made by Nidec and marketed by Scythe. Scythe model D1225C12B9AP-30, rated at 4250 RPM.
First, look at the stock setup. The overclocked CPU is held to a temperature over ambient (TOA) of 51 °C with a sound pressure level (SPL) measured at the side of 25.5 dBA. In other testing I consistently measured the equivalent of 28 dB at the face of the push fan. No matter how you measure it, this is a quiet system. It sets a high bar for cooling and a low limbo-style bar for quiet operation.
Now we can look at the solo center fans. As usual, cooling power improves with fan speed, but higher speed fans are noisier fans. Put simply, in a given system if you want better cooling you must put up with more noise. This is why we choose our heatsinks with such care.
But the main point of this preliminary endeavor is to give us a reference point on what these fans are capable of doing when they are in the center position. A reference chart, if you will.
Seven-Bladed 25 mm Push Fans
We will now try to replicate the stock setup with more powerful fans, starting with seven-bladed specimens. Remember that when you look at the RPM achieved by the push fans, these fans are pushing against resistance so they spin slower. If you want to see the effect of this, take your hand and hold it parallel to any running fan on the output side — a fan in a case will do. As your hand approaches the fan, it will slow down. This will start to happen about three inches (75 mm) away. It’s amazing what such a seemingly insignificant act will do.
So, what have we learned in this round?
Overall, these are excellent results. True, when we include the noisy Thermaltake A 2018 we improve cooling by less than 5 °C and made a lot of noise doing so. If we limit ourselves to setups with the much quieter KM2-1700 in the center, we see an improvement of 2.2-3.4 °C over the stock setup, with SPL ranging from 30-33 dBA. There is a bit of a trend toward lower temps as fan speed increases, but it’s not much.
So it seems as if the KM2-1700 needs help; by itself it does not cool as well as the stock set of fans. But paired with another 25 mm fan it can get another 2-3 °C better cooling with only a moderate increase in noise.That’s a temperature over ambient of less than 49 °C to less than 48 °C. If your ambient temperature is 30 °C (86 °F) your core temp would be under 80 °C.
There are plenty of coolers that never see this territory with noise anywhere near this low. When would you want that extra cooling? If you have high ambient temperatures and want to overclock, you might want it.
The next question is: can we improve these results with fans that use more blades?
Nine-Bladed 25 mm Push Fans, and More
Continuing on our quest to improve on the performance of the D14’s stock fans, we come to the collection of nine, eleven, and sixteen bladed fans. I’m not going to post numbers on the performance of the individual fans. By now you know what to look for.
So what did we learn? Again, the combination of a 1700 RPM Kaze Maru 2 with any high performance 25 mm fan with get you a really good TOA — good enough to really overclock with, even in hot weather. With a spread of 1.7 °C, there is not much difference in cooling here. There was more difference in noise. Probably the CM R4 Sickleflow is the best combination of low-ish noise and good cooling. It matched the best of the 7-blade testing. So, the 25 mm fans that did the best, and most quietly, were:
- Cooler Master R4 Sickleflow
- Cooler Master Blade Master
- Zalman ZM-F3
(I’m not counting the San Ace Silent medium because only fanatics get these.)
Triple Fan Sets
I did another thing: I tested most of these fans in push-pull pairs, making a sandwich of the KM2-1700. I have pairs of both Cooler Masters, all the San Aces, the Gentle Typhoons, the Slip Streams, the Zalmans, the Yate Loons and the Rosewill RFX’s. The improvement in cooling ranged from 0-0.6 °C, at a cost of 1.5-4.5 dBA.
Yup. You read that right. Zero to six-tenths of a degree centigrade is all the additional cooling you get when you put that third 25 mm fan on a D14. So if you have a Kaze Maru 2 1700 RPM in the center and a single 25 mm push fan, save your money and don’t bother with a pull fan. Two fans are enough.
Now, is that still true for setups with 38 mm fans? Let’s find out.
38 mm Fans
There are lots of different ways to display the data from the 38 mm fan testing. I’m going to spare you most of the gory details and give to the summary set. Some of it I can just tell you about; charts are a waste for that.
To start, if you have a 38 mm push fan and a competent center fan, adding a pull fan gets you 0.1-0.7 °C additional cooling. I say ‘competent’ center fan because if the fan is not powerful enough, removing it from the center of a push-pull setup will actually improve cooling: if the center fan can’t keep up it just gets in the way.
Which leads us to 38 mm push-pull setups with no center fan. To boil down the data for you, it amounts to this: if you have a center fan in the same league as your push fan, go with a push fan and a center fan. Otherwise, go with a center-less push-pull setup.
As an example, let us look at a setup for a common fan, the Scythe Ultra Kaze – 2000 RPM (UK2K), which the brand seller claims will push 87 CFM. We will pair it with the Scythe Kaze Maru 1700 which is alleged to push 92 CFM. Since I don’t have two UK2K’s, I undervolted a UK3K to 2000 rpm in free air and named it UK3K.2000 for the test.
The results are interesting. Let’s start with the single UK2K pushing air through the D14. It cools the 4 GHz overclocked system down to a TOA of 50.5 °C, a half-degree improvement over stock at a cost of 4.5 dBA.
Adding a second fan cools it down even more. Now it matches a Slip Stream 1900. The difference here is that at the middle of its speed range, the UK2K is liable to last much longer than the SS-1900, which is stretching the top of its range.
Adding a third fan improves cooling by another 0.6 °C. Removing the center fan tells us that it was helping a lot: the TOA worsens by more than 1 °C; and the TOA is not as good as a single UK2K with the KM2-1700.
Overall, then, the center fan really adds to the cooling, but a third fan does not. Having a push and a pull fan is better than having only a single push fan, but it does not cool as well as having a push and a center fan. These patterns hold true all the way up, until the center fan no longer can keep up with the push and pull fans. That happens in the mid-3000’s, where my most powerful center fan is the 116 CFM AP-30. When I get to that point, with my Mechatronics pushing and one of my Deltas pulling, I get my lowest TOA — 45.2 °C. But the noise it makes — 55 dBA — is too loud.
I expect that the 5400 RPM AP-31 (150 CFM) would be able to keep up with fans that powerful, but be prepared for serious noise.
So let’s get to comparing 38 mm fans. We will look at 38 mm push fans with 25 mm center fans, and 38 mm push fans alone. The results are arranged from lesser to greater cooling. You can see the cost of each arrangement in noise. Note that the entire range in cooling is 1.8 °C.
The fans. Note that except for one, the 38mm fans all have seven blades. Also, most have ball bearings.
|Fan||Amps||RPM||CFM||Static Pressure||Sound Pressure Level|
|Delta AFB1212SHE-4F1A||0.75||3700||151||14.5 mmH2O||55-58 dBA|
|Mechatronics MD1238H12B||1.6||3800||174||16.7 mmH2O||58 dBA|
|Mechatronics MD1238E12B||2.1||4200||192||20.4 mmH2O||59 dBA|
|Nidec CNDC12Z7RP||0.71||3200||124||10.7 mmH2O||49 dBA|
|Nidec GT AP-30 (25mm)||0.56 (1.35 start)||4250||116||9.6 mmH2O||44 dBA|
|Nidec GT AP-31 (25mm)||1.14 (2.69 start)||5400||150||15.2 mmH2O||50.5 dBA|
|Panaflo Medium||0.34||2100||86.5||N/A||35.5 dBA|
|San Ace 9G1212M101||0.21||1950||74||4 mmH2O||32 dBA|
|San Ace 9G1212F101||0.28||2280||87||5.5 mmH2O||36 dBA|
|San Ace 9G1212H101||0.38||2600||99||7.2 mmH2O||39 dBA|
|San Ace 9G1212E101||0.61||3100||118||10.2 mmH2O||46 dBA|
|San Ace 9G1212G101||0.98||3600||137||13.8 mmH2O||49 dBA|
|San Ace 9S-high (25mm)||0.39||2700||86.5||4.6 mmH2O||36 dBA|
|Scythe Ultra Kaze 2000*||0.25||2000||87||N/A||33 dBA|
|Scythe Ultra Kaze 3000*||0.6||3000||133||8.2 mmH2O||46 dBA|
|Thermaltake A2018 (25mm)*||0.48||2800||94||4.1 mmH2O||47 dBA|
* These fans are not sold by the actual manufacturers, so I don’t fully trust their advertised specs.
The 38 mm fans are grouped at around 2000 RPM, mid-2000’s, 3000 RPM, and mid- to high-3000’s. It gives us comparisons between Panaflo, San Ace and Ultra Kaze near 2000 RPM, San Ace -H fans in the mid-2000’s, San Ace, Nidec and Ultra Kaze around 3000 RPM, and Delta, Mechatronics and San Ace at the top. So for the most part we can compare three types of fan in each of our groups.
So what did we learn?
Reading down the chart, you can see that a single 3000 RPM fan cools roughly the same as 2000 RPM fans teamed with the 1700 RPM Kaze Maru 2 (KM2-1700), but makes a lot more noise. Considering the range of TOA in these tests, you could argue that going with the quietest setups gets you pretty good cooling without making your rig really noisy.
At the high performance end of the chart, the UK3K with an AP-30 is among the best cooling solutions. And the UK2K does the same down at the 2000-rpm end. So pretty much this is a vindication of the Ultra Kaze design.
Noctua built the NH-D14 to cool well and do so quietly. It even provides adapters to quiet the fans even further, secure in the knowledge that their heatsink will still do a good job cooling. So why go the other way?
If you live in a hot place, you will want the best cooling you can manage. If you live in a temperate place, you just may want to overclock your CPU.
One reason for doing a survey like this is to map out the parameters of what cooling our little machines can do. In this case, we can see that the last degree or two comes at the expense of great noise. Although it would be interesting to see what 250-300 CFM fans would do in cooling this heatsink, I suspect you would barely crack 45 °C TOA. That would make it a 6 °C improvement over stock fans, with untold noise.
Looking around, I see that the latest Vortez cooler review has the H-100 with four stock fans beating the stock NH-D14 by 2.25 °C. I dare say that with an Ultra Kaze 2000 and a Kaze Maru 2 1700 on a D14 we could beat that cooling, and be quieter while we’re at it.
With various higher performance fans I was able to get my TOA down to 48 °C with only mild to moderate noise. Since the i7 860 runs hotter than most other CPU’s, you should be able to do better than this. Think of running a 4 GHz overclock with a temperature of 30 °C (86 °F) in your room. Think of pushing your overclock to where it is the voltage that scares you, not the CPU’s temperature.
So rev the engine. Pop the clutch. It’s like having an old school V8 under the hood again.