AMD requires liquid cooling for high TDP Threadripper processors, and AMD will not honor warranty coverage for damage and/or reduced performance by using an air cooler. If you're that worried about a leak, get a Corsair, their warranty coverage includes damage to other components caused by a rare leak.
really would be nice to see some real cooler test for 3970x , I'm totally lost what to get. I watched tests between air top tier Noctua fro 1950x threadripper vs AIO and air smashed water. Now for new TR everyone is telling somethign different but no real tests to prove it. CPU is coming next week and I still don;t know what to do about cooling 
Common, Enermax has the worst reviews everywhere (gunk grows inside it), also only 2y warranty compared to 5 corsair or 6 nxtz. Are you Enermax saleperson by anychance ? 
Hello,
I just registered here to provide you with some input on the NH-U14S TR4-SP3. I've been running my new Threadripper 3970X with that very cooler for almost 4 weeks now, and under very high loads as well. Case is a SilverStone Primera PM02. "Very high loads" means all cores and all threads being loaded 70-100%. Sometimes load would be 100% for over 24 hours.
Like this:
As you can see, the load is high enough to fully saturate the entire chip and all of its hardware threads. The code running on the CPU makes use of SSE (1, 2, 3, 3S, 4.1, 4.2), FMA3, BMI1 as well as AVX & AVX2. I linked all fans to the CPU temperature and configured the fans to make use of the "performance" profile in ASRocks' UEFI firmware. Mainboard is an ASRock TRX40 Taichi.
The cooler itself is equipped with two fans, both NF-A15 PWM. The case has a 120mm on top (NF-A12x15 FLX due to space constraints) and a 140mm (NF-A14 FLX) on the rear for exhaust. Intakes are 4 × 140mm, three at the front, one on the top, also NF-A14 FLX.
Then I wrote myself a small script to give me the average temperature across all CPU dies on FreeBSD UNIX 12.1-RELEASE-p1, which I'm running on the box:
#!/usr/bin/env sh
pkgCores=8 # Cores per die
curTemp=0 # Current temperature of a specific CPU die
sumTemp=0 # The sum of all individual temperatures
i=0 # Loop counter
cnt=$(($(sysctl dev.cpu | grep desc | wc -l) / 2)) # Number of physical CPU cores
# Walk through all physical CPU dies and sum up their temperatures
# (There's only one sensor per die)
while [ ${i} -le ${cnt} ]; do
curTemp=$(sysctl dev.cpu.${i}.temperature | cut -d' ' -f2 | cut -d'C' -f1)
sumTemp=$(printf "scale=10; ${sumTemp} + ${curTemp}\n" | bc)
i=$((${i} + ${pkgCores}))
done
# Calculate average temperature over all physical dies
finalSum=$(printf "scale=1; ${sumTemp} / $(printf "scale=10; ${i} / ${pkgCores}\n" | bc)\n" | bc)
# Output
printf "Current average CPU temperature: ${finalSum}°C\n"
I just call it "temp". And this is what it gives me after 2 days of load (2 days, because I've switched the machine off 2 days ago to install some fan grills):
$ temp
Current average CPU temperature: 77.1°CTypically, it's in the range of 75 - 78°C, and sometimes there are spikes where it briefly reaches 82-83°C. Environment temperature at this point was 22°C.
The machine isn't exactly silent anymore when running the fans like this, but it does work without throttling.
Hope this helps.
I've been wondering the same thing just very recently actually, because I want to upgrade from 64GiB to 256GiB of RAM, as I tend to run out of it here and there. Also, I might be able to speed up some of my computations by sacrificing more memory. This'll be especially important when the Zen 3 Ripper arrives, because I'd love to go for 64 cores with higher IPC. 
Where I live (Europe), I can only find one eligible, currently available very-low-profile module, and that's the Micron MTA18ADF4G72AZ-2G6B2, which is a 32GiB DDR4/2666 ECC module. I'd welcome the ECC, wouldn't hurt, but it's a bit slow.
I'll need to check whether normal height modules are really a problem, or whether it's only the sticks with overly large heatsinks that get blocked by the Noctua. As far as I could see with a quick, superficial check, you'd only need to mount the fan(s) a bit higher when you use modules like mine (Kingston HyperX 16GiB at the moment, part no. KHX3200C16D4/16GX).
I'll get back to you with an update as soon as I can make sure.
Heh, nice. Have used the nick for about 20 years by now, so that wasn't intentional. 
Anyway, does the cooler itself block the RAMs, or just the fans? Because you can raise the fans to give more space to RAM modules. If its the cooler itself, then bad luck I guess. I don't need that much RAM, so I only needed 4 × 16GiB modules, more than enough for my workload.
By the way, I've been informed that Noctua is already testing a TR4-SP3 prototype of the NH-D15, so I'm hoping that they will finally bring a "NH-D15 TR4-SP3" to the market in 2020. Would be great for the 3990X as well. The information comes from somebody who has met up with Noctua on the Computex, where they showed a demo of the cooler using a Threadripper 2990WX CPU.
Here's a [link] to the source, but that's all German, unfortunately. Reason is that that person - just like me - is from Austria, like Noctua themselves as well.
I hope your watercooler won't damage your Threadripper rig!
