SFF.Network Noctua presents NH-U12A premium class 120mm CPU cooler

Comment:
We had a quick look at Noctua's new coolers at Computex 2018 - including a prototype for this very cooler. With 140mm cooler performance in a 120mm cooler form factor, this was exciting news for us in the SFF world! Using a denser fin stack, as well as the new generation Sterrox fans, the cooler offers some hefty cooling.

We can't wait to see reviews for this cooler to see if the numbers stack up!

Read the press release here.
 

Wyd4

Cable-Tie Ninja
Oct 21, 2018
196
218
I was absolutely stoked to see this come through the works, and having some of the sterrox fans already, I can vouch for the silence and performance on that front. Though sadly I fear it will still be too tall for most SFF builds?
 

sgtspiff

Caliper Novice
May 25, 2018
30
20
Tall But really good performance.
The ever present dilemma.
"if i just ad a litre or two it Will be both cool and quiet"
 

Bioforce

Airflow Optimizer
Aug 31, 2018
251
116
I like it, but 158mm tall is rough. If you're not expecting compatibility issues with the GPU or RAM, you might as well go the extra 7mm and get one of their top end coolers.
 

Mortis Angelus

Airflow Optimizer
Jun 22, 2017
283
277
I pre-ordered one, mostly because I'm a sucker for fancy air cooling, but also because I'm super interested to see how it stacks up against my NH-D14, DRP4 and Scythe Fuma.

I hope you can do some benchmark testing and post the results here. Perhaps even get the test featured on main page. Would be interesting to see if Noctua can deliver on these bold promises.
 

Allhopeforhumanity

Master of Cramming
May 1, 2017
545
534
I hope you can do some benchmark testing and post the results here. Perhaps even get the test featured on main page. Would be interesting to see if Noctua can deliver on these bold promises.

I'll let you know once it arrives. Unfortunately I don't have any sound pressure measuring equipment to do a noise normalized experiment, but I can definitely do a comparison in thermals vs fan speed.

Someone who got a review sample did some benchmarking: https://pbs.twimg.com/media/D2cFY2kUwAAnjj-.jpg

With a full video here:
Looks like it stacks up pretty well against the DRP4 and NH-D15 with the latter still being a bit better. His tests were in a Meshify-c in the stock configuration, so I'd bet that with some better front fans you could bring the temps down further. I have my DRP4 in a Meshify-c as well, but with 2x 140mm Noiseblocker eloops, so we can see how the delta differs once I get the NH-U12A.
 

Allhopeforhumanity

Master of Cramming
May 1, 2017
545
534
Here's another review from PC Builder's Club, this thing comes really close to the NH-D15.

Seems pretty consistent with the other review I've seen. One thing I've noticed to be lacking is how the extra space between the CPU cooler and the top PCI-E slot affects GPU thermals. When I swapped from a Scythe Fuma to my NH-D14 I noticed a 2-3 C rise. This increase was pretty consistent with the DRP4, so maybe I'll see a few C drop going back to a 120mm design. Amazon has the U12A scheduled for delivery tomorrow, so I should be able to begin testing this weekend.
 

Allhopeforhumanity

Master of Cramming
May 1, 2017
545
534
Alright, so I kind of went down a rabbit hole with my testing trying to find a methodology that would be representative. I had originally planned on testing the U12A against the results I had already collected when I got the DRP4 to replace my NH-D14, but I was never really happy with my temps overall (~92-94 C) with the following setup:
Software: Prime95 29.4b8 small FFTs
CPU: 4790K at 4.5ghz all core, 1.23 V which became 1.26 V after LLC offset and everything else on auto
Mobo: Gigabyte Z97 Gaming G1
Ram: 32 GB of G-skill Sniper PC2400
Case: Fractal Design Meshify C with the solid top from a Mini-C to try and directionalize the airflow
Intake: 2x 140mm Noiseblocker E-loops
Exhaust: 1x 120mm fractal design Venturi HP

One thing I noticed was that the package power on the CPU was all over the place, anywhere from 90-165W and I was definitely getting clock throttling, which I wasn't savvy enough to look into when I tested the NH-D14. So I thought since I was tearing down the PC anyway, I would delid the CPU and start over with the U12A and a full suite of NF-A12x25s (3x as intake and 1x as exhaust).

During the delidding process, right away I noticed that the stock paste between the chip and the IHS was rock hard and based on the pattern it left on the chip was already delaminating. After scrapping off all of the adhesive, coating the SMD resisters/caps in liquid electrical tape and applying the conductonaut liquid metal, I decided not to reseal the CPU to minimize the bond line between the CPU and IHS. I put everything back together with the U12A and reran Prime95 29.4b8 small FFTs with the same settings and my temps dropped to 71 C max after 30 minutes. With such a giant disparity, I couldn't exactly use the previous data on the other coolers as it would be like comparing apples and walnuts.

So then I went down the path of looking at standardizing a testing setup that would be robust and relevant to real world conditions. I bumped the vcore to 1.27 knowing that the offset would stick the voltage right at 1.3 and set the boost profile to be 4.7ghz 2-core and 4.6 4-core and tried locking the boost power to 125 W. Unfortunately this boost power is the most finicky value and the number reported by HWMonitor never got to where I'd set it in the bios. For example, to get 125 W on small FFTs, I had to set the boost limit to 145 W, but then this setup was unstable in large FFTs. So then I tried to aim for 150 W (knowing that auto had it going all the way to 165 W) and had to set the boost value to 180 W. This generated more heat at the same clocks and voltage readings, which must mean that the cores are getting fed additional current, and now the memory controller seemed unstable because I'd get a new blue screen in the blend tests.

So, I bumped the RAM voltage from 1.5 to 1.65 manually because the XMP profile which said that it was increasing the voltage to 1.65 in the bios' memory page still had 1.5 in the voltage page and that's what was being reported by gigabytes system viewer. This seemed to fix the stability issues in blend and large FFTs but was increasing the package power to ~158 W at the boost limit of 180. Thus, I dropped the boost limit to 175 and that seemed to limit the package power back to 150 W.

At this point you must be saying to yourself, "Allhope, you're crazy, why are you so focused on the package power?" and thats because I feel like this is going to be the only representative value to compare coolers in a thermodynamically sound way. CPU Temp (heat stored) = heat generated - heat dissipated, and each cooler is going to behave differently enough as it is with their different number of heat pipes, differently sized interface plates, different fin densities, and different total surface area, that I didn't want to go convoluting things further with unstable input loads. (My gods, this feels like my PhD all over again!)

Anyway, long story short, I've gotten to a place that I'm happy with the setup, but I don't have time for any more testing this weekend. As far as the U12A goes in totally qualitative terms: It is a beast of a cooler for its size.

The following summarizes the tests with the cooler:
CPU: 4790K at 4.6ghz all core, 1.27 V which became 1.3 V after LLC offset and boost power set to 175 W for a reported 150W in small FFTs
CPU thermal interfaces: Conductonaut between the chip and the IHS and Cryonaut between the IHS and the CPU cooler.
Mobo: Gigabyte Z97 Gaming G1
Ram: 32 GB of G-skill Sniper PC2400 locked to 1.65 V
Case: Fractal Design Meshify C with the solid top from a Mini-C to try and directionalize the airflow
Intake: 3x 120mm NF-A12x25's locked at 60% fan speed (~1400 RPM +- about 20RPM variance in the PWM) [100% PWM ~ 2200 RPM)
Exhaust: 1x 120mm NF-A12x25's locked at 60% fan speed (1400 RPM) lined up precisely with the fans of the cooler to minimize turbulence.
CPU Cooler: NH-U12A with 2x NF-A12x25's
Ambient Room Temp: 24 C (measured by wall thermostat)

Software: Prime95 29.4b8 (then corroborated with 29.7b1)
After 15 minutes of small FFTs
CPU Cooler Fan Speed - Package Temp
2000 RPM - 76 C (52 C delta)
1800 RPM - 76 C (52 C delta)
1600 RPM - 78 C (54 C delta)
1400 RPM - 80 C (56 C delta)
1200 RPM - 83 C (59 C delta)
1000 RPM - 85 C (61 C delta)
800 RPM - 89 C (64 C delta)
600 RPM - 94 C (70 C delta)

I don't have any sound pressure measuring equipment, but in my current setup pegging everything at 60% (1400 RPM) is an unobtrusive noise level that I barely noitce, so I set a fan curve ranging from 40% (~900 RPM) at idle to 70% (~1600 RPM) if the CPU ever gets to 90 C.
 

teodoro

SFF Lingo Aficionado
Oct 8, 2018
109
77
CPU Cooler Fan Speed - Package Temp
2000 RPM - 76 C (52 C delta)
1800 RPM - 76 C (52 C delta)
1600 RPM - 78 C (54 C delta)
1400 RPM - 80 C (56 C delta)
1200 RPM - 83 C (59 C delta)
1000 RPM - 85 C (61 C delta)
800 RPM - 89 C (64 C delta)
600 RPM - 94 C (70 C delta)

I don't have any sound pressure measuring equipment...
thanks for this rpm breakdown, it's super helpful. it seems like the only rational way to get an idea of what something sounds like is to have a shared piece of hardware (a12x25 fan) at set performance levels (a given rpm) which is unfortunately rare data to have. I really like the noise profile of the a12x25, to me it's basically inaudible to ~800 rpm and still very quiet to ~1200 rpm. are you planning to stick with this cooler vs. the dark rock pro 4?
 

Allhopeforhumanity

Master of Cramming
May 1, 2017
545
534
thanks for this rpm breakdown, it's super helpful. it seems like the only rational way to get an idea of what something sounds like is to have a shared piece of hardware (a12x25 fan) at set performance levels (a given rpm) which is unfortunately rare data to have. I really like the noise profile of the a12x25, to me it's basically inaudible to ~800 rpm and still very quiet to ~1200 rpm. are you planning to stick with this cooler vs. the dark rock pro 4?

Yeah the noise profile on the A12 is great and as you said, totally silent at < 800 RPM. The silentwings 3 on the DRP4 are really quiet though too, maybe even a touch quieter raw sound pressure wise, but a little more higher pitch in their sound profile.

The loudest component in my main rig is my MSI 980ti gaming X, which is running a few C cooler with the U12A allowing me to reduce the fan curve slightly relative to the DRP4. The U12A (and D14/D15 similarly) have a much nicer mounting system too. The DRP4 uses a separate tensioning bar and it's pretty finicky to get it lined up and level, leaving to uneven mounting pressure if you don't take your mobo out of the case and be really careful in tightening it down. The first time I actually sheered a screw right off, but to bequiets credit they rushed me a replacement mounting system in 2 businesses days directly from Germany.

I liked the all black anodised look of the DRP4 through the tempered glass, but I don't run any RGB and found myself looking at the case less and less, so unless I go with an aftermarket GPU cooler as my next project, I'll probably stick with the U12A for a while, unless my Nh-D14 fits the U12A mounting bracket, in which case I might make a quick swap to see how it performs with 2 A12x25 in the larger fin stack.
 
  • Like
Reactions: teodoro