Cooling Sub-70mm TR4/SP3 air cooler discussion

[Analogue Blacksheep]

TR4 and SP3 being Threadripper and Epyc respectively.

So, I am asking the community, using the current technology avaliable (Copper, Graphene coating, 3D Printing etc) is it possible to make an air cooler for the TR4/SP3 platform that is under 70mm? The smallest cooler out there so far is the Noctua NH-U9 TR4-SP3 at 125mm tall.

Edit - Turns out Dynatron makes 5 of them, click SP3 on the side tab to see them and Supermicro makes 2 of them.

 

[VegetableStu]

i think current heatsinks for SP3/TR4 around that height rely on forced airflow to get things cooled effectively ._.
https://www.dynatron.co/product-page/a34

if you want a U9 but under 70mm tall, the next nearest fan would be 60mm in diameter
https://www.dynatron.co/product-page/a26-1

 

[Analogue Blacksheep]

I was asking because I've decided I will be doing my build in a Cemo M1. I didn't consider Dynatron would have something. If the a26 was combined with a Noctua NF-A6-25 PWM would it be better?

 

[VegetableStu]

to be completely honest... I have them right now as a fallback cooler. I'm just missing a threadripper system, so I can't test how much better or quieter it cools ,_,

EDIT: although I should mention the fan bundled with the A26 is rated for higher speeds and static pressure, so it'd be a better cooler flat-out but i bet it'll be screaming. might need a steep curve for that on idle ._.

 

[Analogue Blacksheep]

I just looked at the A26 fan. At 20% it's at 1600 RPM, 50% is 4000 RPM and 100% is 8000 RPM. At 100%, the Dynatron fan has an airflow of 43.75 CFM (74.33 m³/h), static pressure of 19.04mm-H2O and noise level of 53 dBA. The Noctua on the other hand hits 3000 RPM. At this speed it has an airflow of 17.19 CFM (29.2 m³/h), static pressure of 2.18 mm H2O and a noise level of 19.3 dB(A).

 

[VegetableStu]

i did a quick excel maths and they're right about the same (CFM at 3000RPM should be about 17.40405, while the NF-A6 does 17.186471 CFM (used an online converter).
noise goes to noctua (31.6613 dBa vs 19.3 dBa), but realistically I'd question cooling the thing at load ._.

 

[Analogue Blacksheep]

Say hypothetically you have a the CEMO M1 or the L1. You have a Asrock X399M board in it with the Dynatron A26 pointed towards the two 80mmx15mm fans (Noctua should be bringing out slim 80mm later this year). The hot air from the CPU is going to be pushed very quickly out of the case. Might that work?

I feel like we are at the unexplored frontiers of SFF right now, here is an example build that could go into the CEMO M1.

FYI, the Dynatron A26 has a TDP of 180W.

 

[VegetableStu]

depends on how Dynatron tests their heatsinks I think o_o they didn't mention anything other than it being done at ambient temperatures of 35c.
although one could argue the Cemo case would be closer to open-air, just because the inlet and outlet apertures are pretty close to the walls of the case, but that's pure speculation ._.

if I had the blowaway budget that's enough for me to feel adventurous, I'd actually try using an NF-A12x25 above the copper chamber Dynatron A28 o_o
https://www.dynatron.co/product-page/a28
the stack should be about 52mm tall
but like you said: it's not something that has been experimentally documented yet ,_,

... I might want to add more holes to the side panel though

 

[Valantar]

My partner's 1920x is cooled by an NH-U14s with two fans which at best avoids throttling under prolonged all-core loads like rendering (according to HWMonitor it hovers around 175W package power draw). That is at bone stock settings, which of course means a very low throttling temperature for Threadripper - 68C, IIRC. Can't remember if this includes any offsets or whatnot - due to it being a rendering rig and not mine, I haven't wanted to mess too much with it.

I can't imagine a fin stack a quarter the size of that could possibly keep up unless fan speeds were increased to match. Those 80mm Noctuas described here sound woefully insufficient.

 

[VegetableStu]

(according to HWMonitor it hovers around 175W package power draw)

hmm that's about above 180W in heat output i think? i don't have a conversion table for this ,_,

 

[Analogue Blacksheep]

I found this article on Phoronix which looks at the Dynatron A26 and compares it to other coolers. Bear in mind the CPU is the 2990WX which is a 250W part.

I was thinking, what if there was a way of installing a second fan on the A26?

I also found another two coolers to add to this crazy group, the Supermicro SNK-P0063AP4 and the Supermicro SNK-P0063P.

 

[VegetableStu]

adding more heatpipes will help more than adding more fans at this point ._.
also don't forget such server coolers are installed in a super high airflow environment (i.e. the server cases have lots of air forced through it with many high RPM fans)

 

[VegetableStu]

if I had the blowaway budget that's enough for me to feel adventurous, I'd actually try using an NF-A12x25 above the copper chamber Dynatron A28 o_o
https://www.dynatron.co/product-page/a28

seriously, I'd look into that, LOL. it's the exact base cooler as the one Linus was using in his first Dan A4 build (minus mounting differences and model names) and it's pretty much within spec. if that base cooler has a higher TDP rating without major changes I'd wager it'll scale better with more airflow

(... uhh media embedding not working...? youtube link )
https://www.dynatron.co/product-page/t318

 

[Valantar]

I found this article on Phoronix which looks at the Dynatron A26 and compares it to other coolers. Bear in mind the CPU is the 2990WX which is a 250W part.

I was thinking, what if there was a way of installing a second fan on the A26?

Remember that a second fan in-line only increases static pressure, not airflow (other than helping alleviate airflow blocked by a restrictive fin stack). Performance increases would then likely be small, however they might be noticeable given that you're speculating in switching out a high-performance 8000rpm fan with a low-noise 3000rpm fan. Still, given the design of the heatsink the only way to add a second fan is likely making a duct off the end without the fan and mounting a fan to the duct. I sincerely doubt this would be worth the effort. If you look at the spec sheet for the A26, they only list its thermal perforamance at fan speeds between 6000 and 8000 rpm. Server heatsinks like these are not designed to run at low fan speeds even if the fans are technically capable of it.

An A28 with a larger fan sounds far more likely to work, but even that sounds doubtful given that it's designed for use in 1U servers with massive arrays of >10000rpm 40mm fans. If you ducted the fan so that all airflow went through the fin stack, that would help, at least.

 

[Analogue Blacksheep]

Yeah, I'm starting to think this might be a lost cause for the time being. I've watched a video from @Josh | NFC that showed an i9 7920X (A 140W part) in an S4 Mini and he struggled to get it cool. Plus there are rumours of an 16 core Ryzen which may make this whole idea irrelevant.

Thanks @VegetableStu and @Valantar. My curiosity has been stated. Let's see if the future will bring someone else crazy enough to try this.

 

[fedder]

Not to Necro this but I've been wondering. Anyone think it might be possible to use the Dynatron A28 but with the Blower from the A31 mounted onto it for a but more cooling? It probably wouldnt help too much but its hard to say. I really wish more proper cooler options existed for TR4/ SP3.

Incase anyone is wondering im thinking of shoving a Threadripper with the PPT turned down into the ASrock TRX4 Creator with a bunch of blower gpus into a Cerberus X. The A28 with a blower modded onto it or a A31 would be the only realistic options for aircooling.

 

[Analogue Blacksheep]

I had some crazy ideas back then. I think one of the answers that might arrive are Thermalsiphons. Imagine a low profile one.

 

[jpmomo]

I have built a few sff PCs using the dynatrons and noctua directly on top blowing down. I have done this with a 28 core Xeon platinum and both 7452 and 7742 epyc Rome CPUs. It actually does a good job of keeping the thermals under control with a low total height. I used the noctua a12x25 and their industrial 140 3k rpm beast.

 

[thelaughingman]

I had some crazy ideas back then. I think one of the answers that might arrive are Thermalsiphons. Imagine a low profile one.

yes, but they need to get the standard version out of the gate first. imagine the possibility if they make a low profile version, mouth watering indeed!