Noctua TRX4 Threadripper Cooler Roundup

The Noctua TR4-SP3 coolers in this review are designed to fit AMD socket TRX4/TR4/SP3 processors. Here is an overview of the system and testing methodology.

TRX40 Aorus Master – TRX40 Chipset
AMD Ryzen Threadripper 3960X (3.8Ghz) 24 Core 24 x 512KB L2 Cache 8x 16MB L3 Cache

Coolers and Variations
NH-U14S TR4-SP3 (single NF-A15 fan)
NH-U12S TR4-SP3 (single NF-F12 fan / Dual NF-A12x25 fans)
NH-U9 TR4-SP3 (dual NF-A9 fans)

The CPUID System Monitor was used to obtain and record system temperature data and being that this is a quad core processor we need something that will work across all of the cores at once.  For this task we're using a new version of Prime95 (p95v2810) that will allow you to spawn (n) instances to test with.

Editors note: Even though the Windows 10 task manager reported 100% processor usage we could never attain a 100% of the rated heat output as documented by AMD when using Prime95 as a basis for that heat production. Knowing this we ran the stress test until the maximum temperature was attainted and stabilized.

Other things to consider when judging software induced heat output.
a) Clock throttling by the processor at high temperatures.
b) Normal software isn't designed to produce maximum heat output.
c) Variances of cooling temperature.
d) Variances in CPU load.
e) Inaccuracies in thermal diode readouts.
Of course the list goes on..

Our testing methodology is aimed to provide a real world look into this heatsink given the test system provided.

A C/W rating can quickly be calculated using this formula.

C/W = (CPU temp - Ambient temp)/(Variance(%) * CPU Watts)
Allowed variance for this test = 85%
CPU Watts = 280W

NH-U14S Single
0.26 C/W = (85C - 23C)/(.85(280W))

NH-U12S Single
0.30 C/W = (95C - 23C)/(.85(280W))

NH-U12S Dual
0.29 C/W = (91C - 23C)/(.85(280W))

NH-U9 Dual
0.30 C/W = (95C - 23C)/(.85(280W))

In our heatsink and waterblock tests we don't really focus on overall load temperatures but rather how well the product can remove heat given a specified heat load. Since this is a real world testing method we need to take into consideration real world variables and estimate tolerances. This is why we normally only apply 85% of the total wattage output to our heat calculations.

The resulting C/W number is used to rate how efficient a heatsink or waterblock is based on the given heat load. These numbers can be used to determine heat capacity, the larger the difference the less efficient the heatsink is. (aka not good for overclocking)

With Threadripper we elected not to do any overclocking tests as every cooler would have reacted differently.  It should also be noted that while the test started with several CPU cores running at 4.1Ghz by the end of the test the entire CPU had throttled down to 3.9Ghz.  This throttling amplified when the CPU was overclocked and several cores would run even slower once the CPU reached 95c.

I did find it interesting that my “new” Prime95 test had two sets of load numbers depending on the FFT and loop unlike the previous 32bit version.  Using Small FFTs seemed to make things more even though wasn't perfect.  Sadly I couldn't use the old version due to a threadcount limitation.

Keep in mind these calculations are provided for demonstration purposes only and may not reflect the actual lab tested C/W rating, but we're pretty close.