News Noctua SFF Cooler Roundup: Testing Small Coolers Against Intel’s Core i9-13900K

[Admin]

We tested Noctua’s compact SFF coolers with Intel’s i9-13900K to find out their thermal dissipation limits.

Noctua SFF Cooler Roundup: Testing Small Coolers Against Intel’s Core i9-13900K : Read more

 

[Amdlova]

Wow great review too many noctuas in one review.
You can test the thermalright coolers cheap as he'll
Thanks man!

 

[edzieba]

While all the coolers we tested here were able to handle Intel’s Core i9-13900K to some degree, you’ll probably want to stick with lower-end CPUs if opting for more compact coolers. Because even if your cooler can technically move a high amount of heat away from your CPU, that heat still needs to escape from your case quickly, or it’s just going to be heating up your other components. And if your case requires one of the smallest SFF coolers, there’s a good chance its internal airflow is also likely to be limited compared to larger cases.

With the exception of 'fashion' cases (e.g. acrylic cubes or designs intended to hide all ventilation behind cosmetic panels) an SFF case will usually fare far better than a full size case here. When you have such a tiny volume of air to vent and such small distances between component fan and external intake, case volume changes per minute can be pretty ludicrous. On top of that, intake ducting can net you some serious thermal benefits for very little by turning every component fan into a direct intake with no recirculation, effectively matching performance with an open-air test rig.

 

[Phaaze88]

You can test the thermalright coolers cheap as he'll

This could be Thermalright's fault more than the reviewer's, so perhaps the pitchforks and torches should be raised at them instead?
If requested, I'm sure Noctua has no problem sending out review samples.
As for Thermalright, I'm not sure what they're doing, but it may be connected to the lack of available reviews in english: https://www.techpowerup.com/reviewdb/CPU-Coolers/Air/Thermalright/

Plus, if a reviewer has to purchase a sample with their own cash, they'll be less likely to do it, especially when other samples are on the way, or in their hands already.

 

[Albert.Thomas]

No pitchforks needed, folks. I do have contacts with Thermalright, I actually just finished their Phantom Spirit 120 on my Ryzen system, but there's only so much I can cram into a single review.

I might look at SFF again in the future, but I don't want to continue to try working with the motherboard I have on hand with it's extra tall heatsinks - too many compatibility issues. If I do, I'll probably use my 7700X instead of the i9-13900K, any recommendations for AM5 mITX motherboards?

 

[Phaaze88]

No pitchforks needed, folks. I do have contacts with Thermalright, I actually just finished their Phantom Spirit 120 on my Ryzen system, but there's only so much I can cram into a single review.

Hey, good to hear!
Toms and TPU are 2 sites I frequent, and there would be at least one comment in the cooler reviews requesting more Thermalright, as well in Youtube comments.


I wish they had a larger presence in the US. Their availability and support is not quite there yet.
Only available through Amazon and Newegg now; other sites have dropped off the radar.
Thermalright Direct 1/2/3 is the seller. That may have changed since the last time I looked through.

If the packaging doesn't come with the bracket you need for a cpu, can you request the correct one shipped to you free of charge, or will you be directed to purchase it instead?

 

[Giroro]

For what it is, I like the NH-L9x65. But it should be priced below $30, not $60.
Once you're in the $60+ price range, a small AIO is going to be a better choice for most people.

 

[helper800]

I might look at SFF again in the future, but I don't want to continue to try working with the motherboard I have on hand with it's extra tall heatsinks - too many compatibility issues. If I do, I'll probably use my 7700X instead of the i9-13900K, any recommendations for AM5 mITX motherboards?

ITX boards all tend to have tall surrounding heatsinks. If I had to pick one ITX board, maybe the Gigabyte B650 Ultra. It seems like a decent board, but I don't know if you need any of the extras that come with the X670E, or the B650E boards.

 

[bit_user]

Thanks for the roundup! I know it's on the larger side, but I think it would've been nice to include the NH-C14S as it would round out their selection of down-draft coolers. Plus, the height isn't so bad, if you mount the fan underneath - only 5 mm taller than the NH-D9L!

​

NH-C14S

Successor to the award-winning NH-C14, the NH-C14S is an elite class top-flow CPU cooler that is extremely efficient, highly compatible and remarkably adjustable. Thanks to its deeper fin stack and the renowned NF-A14 PWM fan, the single fan NH-C14S provides similar quiet cooling performance to...

noctua.at

​

I'm personally very interested in it.

 

[greenreaper]

Noctua has the NA-FD1 as well to provide a custom-sized duct to a mesh case intake - in my experience it was good for another 5-10 W of cooling on a 7600 in a Chopin Max case.

They will even send you some bolt extenders if you want to put a bigger 25mm fan on the cooler with the duct (though I suspect it also has to be about as fast RPM-wise to gain any benefit).

I put another fan outside the mesh pointing into the duct to raise the main fan RPM for another ~5 W.

 

[Amdlova]

The noctua thing is made for enthusiasts people who love to hear the butterflies. Thermalright in other hand will attract the people who wants build the first PC cheap build... (new public to the site). Easy to find a thetmaltight on sub 500 build than a noctua one grand build. (Yeah and pitchfork the thermalright (where is the new ultra 120 extreme cooper edition)

 

[StutteryG]

We tested Noctua’s compact SFF coolers with Intel’s i9-13900K to find out their thermal dissipation limits.

Noctua SFF Cooler Roundup: Testing Small Coolers Against Intel’s Core i9-13900K : Read more

Did you encounter any issues at all with the NH-L12S on your Aorus B660I? I was surprised to read that you said it fit without issue since I've read from several sources that the motherboard heatsinks prevent it from fitting properly. Appreciate any insight you can share!

 

[Albert.Thomas]

Did you encounter any issues at all with the NH-L12S on your Aorus B660I? I was surprised to read that you said it fit without issue since I've read from several sources that the motherboard heatsinks prevent it from fitting properly. Appreciate any insight you can share!

It only fit like this, touching the NVMe heatsink

 

[Albert.Thomas]

ITX boards all tend to have tall surrounding heatsinks. If I had to pick one ITX board, maybe the Gigabyte B650 Ultra. It seems like a decent board, but I don't know if you need any of the extras that come with the X670E, or the B650E boards.

The problem in particular was the NVMe heatsink right underneath the CPU. It's much taller than normal NVMe heatsinks.

 

[bit_user]

@Albert.Thomas you've hit on a subject of some interest to me. I've seen data like your Raptor Lake CB23 vs. Power graph, but the problem with presenting it as a bar graph is that the power limits aren't evenly space. It makes the data look rather linear, but it's not.

"There will be some performance loss compared to a cooler that can handle the full heat of Intel’s i9-13900K, but it’s not as much as you might think. I’ve previously tested power scaling on my personal blog, shown below, to see just how much power Intel’s i9-13900K really needs."

​

So, I've taken your data and plotted it on a (X, Y) axis. You can see that it's much more parabolic (or whatever you call a parabola on its side).

​

Since I had trouble locating your personal blog, I got the 315 W figure for "Unlimited" from this article: https://www.tomshardware.com/reviews/deepcool-lt720-aio-cooler/2

I've then added threshold lines for the max cooling limit of each cooler, so you can eyeball about how much performance each should allow. I guess a step further would've been to actually intersect them, but since it's just a rough estimate, I think eyeballing it is fine. We can clearly see that the weakest cooler - the NH-L9i-17xx allows only about 3/4ths of the i9-13900K's performance, while the strongest of this lot (i.e. the NH-D9L) reaches about 7/8ths.

P.S. I like what you did with the noise-normalized results! The only thing is it's going to look different based on where you put the threshold. Still, one plot like that is nice to see and probably helpful for some.
: )

 

[Albert.Thomas]

So, I've taken your data and plotted it on a (X, Y) axis. You can see that it's much more parabolic (or whatever you call a parabola on its side).

​

​

Since I had trouble locating your personal blog, I got the 315 W figure for "Unlimited" from this article: https://www.tomshardware.com/reviews/deepcool-lt720-aio-cooler/2

I've then added threshold lines for the max cooling limit of each cooler, so you can eyeball about how much performance each should allow. I guess a step further would've been to actually intersect them, but since it's just a rough estimate, I think eyeballing it is fine. We can clearly see that the weakest cooler - the NH-L9i-17xx allows only about 3/4ths of the i9-13900K's performance, while the strongest of this lot (i.e. the NH-D9L) reaches about 7/8ths.

P.S. I like what you did with the noise-normalized results! The only thing is it's going to look different based on where you put the threshold. Still, one plot like that is nice to see and probably helpful for some.
: )

That's an interesting way of displaying the performance offered by the SFF coolers, I like it. If I end up doing another round up like this, I might try putting together something like that.

The blog post referred to in the article is :

Intel i9-13900k Power and Cooling Scaling Overview

In gaming, enforcing a 95 watt power limit results in a 5.7% performance loss. More intensive workloads will cause the decifit will increase, in Cinebench imposing a 95 watt limit results in a 30% …

www.boringtextreviews.com

 

[hotaru.hino]

That's an interesting way of displaying the performance offered by the SFF coolers, I like it. If I end up doing another round up like this, I might try putting together something like that.

The blog post referred to in the article is :

Intel i9-13900k Power and Cooling Scaling Overview

In gaming, enforcing a 95 watt power limit results in a 5.7% performance loss. More intensive workloads will cause the decifit will increase, in Cinebench imposing a 95 watt limit results in a 30% …

www.boringtextreviews.com

The only thing I wish the article did was say how long each Cinebench run lasted, and the score doesn't make it obvious what that is. Power consumption needs a time component to get the metric that matters in all this: how much energy did it take to do the thing?

Like the 45W figures may look nice from an efficiency standpoint, but if it takes over three times as long to complete the task compared to max power, it's actually worse.

 

[bit_user]

That's an interesting way of displaying the performance offered by the SFF coolers, I like it. If I end up doing another round up like this, I might try putting together something like that.

You're free to use that image in this article (or elsewhere), if you like, since it's your data. I just updated my post with a slightly improved version.

Charts like this can take annoyingly long to make. So many little details to tweak.
: /

 

[bit_user]

The only thing I wish the article did was say how long each Cinebench run lasted,

Yeah, not having ever run CineBench, myself, I wonder if it's fixed in the workload size or is it fixed in the amount of runtime?

And just what kind of runtime are we talking about? Is it significantly longer than the turbo boost limit? If not, then turbo is going to be skewing the averages, which will have some impact on the applicability of these figures to longer-running jobs.

Power consumption needs a time component to get the metric that matters in all this: how much energy did it take to do the thing?

Sometimes. That matters for workloads of predefined size, like I guess rendering an image or a short clip would be.

However, it's not as if someone switches off their PC, the instant a compute job is finished, nor do they probably have a huge batch of work for it to process, one after another. So, it's a little bit artificial to compute the amount of Joules used to do a chunk of work and compare that from one machine to another. I have a couple ideas about how to account for this, but I'll have to think about it, some more.

Like the 45W figures may look nice from an efficiency standpoint, but if it takes over three times as long to complete the task compared to max power, it's actually worse.

You have to view energy usage in conjunction with performance. For a machine someone is using interactively, you can't afford to look only at efficiency.

 

[TerryLaze]

Yeah, not having ever run CineBench, myself, I wonder if it's fixed in the workload size or is it fixed in the amount of runtime?

The standard test renders one single frame with as many tiles as there are threads and gives a score as a result.
There is also a stress option that will run continuously for 10 minutes.

And just what kind of runtime are we talking about? Is it significantly longer than the turbo boost limit? If not, then turbo is going to be skewing the averages, which will have some impact on the applicability of these figures to longer-running jobs.

If a CPU manages to render the frame in less than one minute which is the max TAU for most CPUs then that alone says a lot for the CPU.
But ever since PL1=PL2, nobody uses TAU anymore, heck nobody uses any power limits anymore.

 

[jayjr1105]

Would like to see the Silver Soul 135 against the NH-D9L

 

[rapidwolve]

It only fit like this, touching the NVMe heatsink

If you think that's tight, Albert, try finding a non AIO for 1 of the Asus ROG Strix i boards ie the Z690I Gaming WiFi.
The VRM heatsink, and daughterboard combination, make trying to find an air cooler almost impossible. I opted for the NH-U9S and it seems to be doing ok as far as a 13600K since I am not overclocking the chip. Anymore and probably have to resort to an AIO.
As for Thermalright, yes they need to step up their gameplan as far as compatibility. (they aren't the only 1's either) That and their customer service needs work. Got a Thermalright Peerless Assassin given to me for my AMD build and found the fans noisy. Sent them the photos, as requested, to show the system and have heard crickets.

 

[cyrusfox]

Fascinating, I really like your maximum watts cooled chart and would love that to include more than these 5 CPU coolers. Also shocking how bad the NH-L12 Ghost S1 Edition performs compared to the Chromax NH-L9i-17XX... We are talking 66mm vs 37mm, this is exactly the reason not many are using Noctua in the height constrained SFF builds, not many choices and they aren't very compelling

I built a 13900 in a K49 case a couple months ago, this limited my CPU heatsink choices to a cooler sub 55mm height, I think for future articles you should zero in on the height constrained choices, although it is great to have these taller options in as a reference to see what you are leaving on the table by going thin. These are the sub 55mm coolers I was looking at:

• Alpenfohn Black Ridge (47mm)
• Cryorig C7 (47mm, copper and graphene versions)
• ID Cooling 47K (47mm)
• Thermalright AXP-90 (47mm, copper and aluminum versions)
• Thermalright AXP-90 X53 (47mm, copper[called Full Black] and aluminum version)
• ID Cooling IS-60 (55mm)
• AXP-100 Copper (with some pushing can be brought down to 55mm)

I ended up going with a AXP-90 copper version as it was pretty reasonably priced as well as being install compatible with LGA1700. Not apples to apples as I am using non-K but I currently boost at 200W but only for 2 seconds then cpu drops to 65W, so I never reach thermal throttle. As such not sure where it ranks but for encoding/gaming/photo work it feels just as fast as the Node 304 it replaced(which was using a noctua nh-u14s). I idle at 55°C, at 65W continuous use all cores stay sub 70°C (Room is 25°C for reference).

 

[bit_user]

I idle at 55°C, at 65W continuous use all cores stay sub 70°C (Room is 25°C for reference).

High idle temps, though. When I built a Haswell i3 (smaller die, worse TIM), it bugged me to see it idle so high.

 

[cyrusfox]

High idle temps, though. When I built a Haswell i3 (smaller die, worse TIM), it bugged me to see it idle so high.

I haven't had time to play with it much and try and undervolt, or do a foam mod, but my idle use is really typing/light browsing + Youtube video playing. Room for improvement but not sure I care enough to play around with it. When the system is not in use I put it to sleep, and with active continuous use below 70°C, it's better than I hoped for. I upgrade often so good chance it will be short lived goin to either arrowlake or meteorlake if it comes to desktop.