Noctua's NH-U9 TR4-SP3 Is Still The Best 4U EPYC / Threadripper Cooler I've Found

[phoronix]

Phoronix: Noctua's NH-U9 TR4-SP3 Is Still The Best 4U EPYC / Threadripper Cooler I've Found

If you are in the market for an AMD Ryzen Threadripper or AMD EPYC heatsink that fits within 4U height requirements, the Noctua NH-U9 TR4-SP3 is still easily the best option available. I'm now running the NH-U9 TR4-SP3 in five different EPYC/Threadripper systems in the racks and they work out splendid...

http://www.phoronix.com/scan.php?page=news_item&px=Noctua-NH-U9-TR4-SP3-2019

[audir8]

Anyone know why TR4 CPUs have max temps of 68C instead of 95C for Ryzen?

https://www.amd.com/en/products/cpu/...#product-specs
https://www.amd.com/en/products/cpu/...#product-specs

Something about MTBF and 24/7 usage for a server CPU perhaps?

Good information in the article for those still salivating over getting a TR.

[Brisse]

audir8 Possibly. Could also be that they messure temperature differently and that the hotspots in a Threadripper messuring at 68°C might actually be the same as the Ryzen's while it is messuring 95°C. Also, some Ryzen's had a 20°C offset to their temperature reading to alter CPU fan behaviour, although some motherboards remove this offset. My ASUS PRIME X370-PRO was like that when it first released but it made it impossible for me to set the fan curves the way I wanted with a NH-D15 so I contacted ASUS about it and asked them to remove the offset which they agreed to and every BIOS release since has not had the offset. Don't know if Threadripper has some sort of offset. Let's just say that temperature messurement on AMD processors are confusing and might not be what it first seems and people end up comparing apples and oranges.

[TumultuousUnicorn]

Michael: Is it possible to turn the cooler of 90° in anticlockwise direction on this model? All heat go through PSU in this configuration...

[baxeno]

Could you run the timed kernel compile a few more times as it doesn't look like the temperature has reached steady-state?

[audir8]

Brisse The packaging would matter, and you're right the way AMD measures temperature might be different on a TR vs Ryzen. It's possible lm sensors reports Tdie(s) and Tcpu. It would also be good to know if there was any thermal throttling being done like in Radeons and most Intel CPUs.

[Alliancemd]

That's some really horrible air-flow path there... Not possible to turn the cooler around so that it doesn't blow into the PSU and instead pushes further the air from the front -> out of the back?

[microcode]

Brisse The packaging would matter, and you're right the way AMD measures temperature might be different on a TR vs Ryzen. It's possible lm sensors reports Tdie(s) and Tcpu. It would also be good to know if there was any thermal throttling being done like in Radeons and most Intel CPUs.

I just checked on my 2950X, and it seems that lm_sensors reports 70° as "high" for Tdie, but does not report any "high" number for Tctl, and unlike my Xeons, the Threadripper does not report a "crit" temperature. I'm not exactly sure about why there's such a difference, though it doesn't seem hard under normal circumstances to keep my Threadripper below 70° Tdie.

Edit: I suspect this difference has to do with the packaging. They can not expect the same maximum local temperatures to be kept from a higher Tdie. I suspect higher power densities will become feasible as they find better measurement points (to more closely follow local temperatures which are the major concern for durability).

[microcode]

I'm currently using the NH-U12S TR4-SP3 with dual NF-A12x25s on my 2950X, and it keeps it nice and cool. the NF-A12x25 is basically silent up to 80% duty cycle, so I never really hear the thing spin up too much except on boot (when the firmware runs everything at 100% duty cycle until the fan controller is started).

[anarki2]

How about noise levels? I'd be curious how it fares against the Dark Rock Pro TR4.