Silent/quiet CPU cooler options for Epyc 9005 (sp5)? (not seeing ANY options - maybe make a custom one?)

Do you know any silent/quiet CPU cooler options for 9005 (sp5)? The goal is for the server (full ATX tower) to be usable in a bedroom.

Noctua has zero offerings.
Silverstone has xe04_sp5, but it’s 43 dBA, which is loud.
Supermicro and Dynatron coolers are even louder.

I would love something in the 30 dBA range. Seems like the few offerings are (understandably) targeted towards the pro server niche where noise is not of concern.

Would swapping a fan in this:

Or adding a fan to this massive heatsink:

…be a viable solution? Does anyone have experience with creating a custom cooler this way: how would you affix and secure the Noctua fan to the 3rd party heatsink?

Thanks

Silverstone currently makes the most performant SP5 air cooler:
https://www.silverstonetek.com/en/product/info/coolers/xed120s_ws/

If you swapped a noctua fan onto the dynatron j12, I’d expect the the cooling capacity to go from it’s stated 580w @ 20c ambient to around 120w max, It’d basically make the cooler unusable.

The J7 is in an even worse situation due to it being so small, but if you had to you could 3d print an adapter shroud for an appropriate fan.

Liquid cooling is pretty much the only way to get a very quiet experience with higher wattage SP5.

5 Likes

Which exactly? They range from 125W to 500W

1 Like

You could use the XE04-SP5, but swap out the fan for a Noctua 92mm fan? That could be something you could try though I don’t know how that would affect thermal performance. If you’re going for a full ATX tower, you can also try the Silverstone XE360-SP5, but swap out the fans for something more quiet.

Not quite the same, but I use an XE360-SP6 and I just took out the fans for something more quiet in a push-pull configuration. No loud noises since.

1 Like

@JayVenturi posted the best resource for cooling EPYC sp5 socket on the internet, he just graced us by posting it on this forum

2 Likes

9115

45 dBA is loud for a bedroom

The ones that I saw were also in the 45 dBA range.

That’s with the cooler going at full tilt, a 9115 under load wouldn’t cause it to spin that fast; however many server motherboards will spin the fans to full speed while booting so you would be exposed to that 45dBa during boots.
To make the cooler even quieter you could swap the fan out for Phanteks T30 in advanced mode and have it run 27dBa max.

Er, 27 dB(A) is perf mode. Advanced is speced at 40 dB(A). Those are free air marketing specs, which are often optimistic (HWBusters got 44 dB(A) in free air), and load noise in actual use at low to moderate RPM often adds a dB(A) or two as well. However, 3000 RPM in free air’s getting to where there’s enough on application spin for the drag to slow the fan down enough to drop a couple dB(A).

Also, assuming all noise is fan noise isn’t necessarily wise. Depending on cooler design, the dominant noise source can be airflow over the fins rather than over the fan blades. In which case fan swapping may not do anything or even make the cooler run louder. So far as I know nobody’s tried to do this attribution for an SP5 cooler.

whoops, you’re correct, that should be set into Performance mode to cap it at 27dBa.

Thats a good point about the fan manufacture’s optimistic specs, however all the fan manufactures are guilty of it, Noctua NF-A12x25’s "22.6"dBa is just as loud as the Phantek T30’s "27.3"dBa in hwbuster’s test.

What’s the problem in buying Dynatron coolers that you posted and swapping fans for Noctua? This is what I do in 2U and 4U chassis where noctua doesn’t have an integrated solution with horizontal airflow but their fans still work great with 3rd party heatsinks.

very poor performance out of the noctua fan, it wouldn’t even be able to move a quarter of the air through the heatsink as the stock fan. It’d probably be okay at idle, and maaaybe even be okay for the low watt skus like the 9115 if you live in a cold place, but there are better choices/options if you have more than 2/3U to play with.

What are the options, I’m curious to learn.

So far I’ve been running AMD 4462P in 2U and 7402P in 4U chassis - I don’t hammer than at 100% but at idle to 10% Noctua fans working fine. I have 2 fans per heatsink on each of them, plus 2-3 bigger case fans that create horizontal airflow pressure.

The best aircooler options for SP5 if you have the height are the dual tower designs from silverstone and coolserver (harder to get in NA atm); both use thicker than “normal” 120mm fans for a better noise profile over the 80mm dynatron.

Another thing to look out for is inadequate VRM cooling from these quieter coolers. I had to put dedicated VRM cooling fans on my server board because I didn’t have the airflow over it that it was designed for since it’s basically a deskside workstation now.

1 Like

Do you have experience doing this swap? Is it a straightforward drop in replacement?

Is it different from swapping for Phanteks T30 in Silverstone due to Silverstone’s heatsink size or Phanteks’s CFM? (or both?) I am trying to understand the science why one is bad and the other is good.

Thanks!

I am building inside Fractal Define 7 XL. It’s a huge tower. Is it wide enough?

Could you please link to the Coolserver one? Maybe they’ll ship to NA. Also, which one is better: Coolserver or Silverstone?

Thanks!

Comparison of Airflow between Dynatron and Noctua NF-A8 PWM fans

image

image

its quite a bold claim that stated max power 580W will go down to 120W

How do you mathematically calculate this exactly ? Surely heatsink takes some of that heat away and fan just helps to dissipate it off the heatsink and its not a linear dependency.

You can always attach 2 Noctua 80mm fans to J12 heatsinks (front + back).

Also, in 4U cases you have additional horizontal airflow from the chassis fans.

Sure, it wont hit 580W but will be close to it.

Also, no one runs those 100% busy in the bedroom.

The noctua is making the same airflow at 100% speed as the dynatron is at 0% PWM signal speed.
This is my paranoid side speaking, but I wonder if noctua are obfuscating their poor airflow performance through using weaker units that are written as larger numbers while everyone else uses CFM.

That 120w figure was a spitball answer based on the dynatron moving ~4 times the air at a higher pressure, not a scientific basis.

Looking through dynatron’s actual performance data they do show this:

There are some problems (no ambient temp given; revisioning down the max TDP to 320w at a different ambient in later datasheet (but not on main page site)) with the data but we can side step them for the most part. If we use something like a power series we can extrapolate a thermal resistance of ~0.13C/W for the 0% PWM RPM performance of the cooler which should be comparable to the noctua’s performance at 100% speed (if datasheet is to be believed).
Assuming all this, the J12 with noctua fan should be able to cool 164.923 watts while keeping the same thermal criteria as the original j12 cooler.
This would be the upper limit, I wouldn’t be comfortable having no thermal headroom with my cooler, so 120 watts is actually a pretty reasonable limit if a little conservative.

It would be a little difficult to attach two noctuas to the j12, the back of it doesn’t have the nubs to attach a fan to that the front does, and has cutouts that would “leak” air, but this could be overcome with an elaborate 3d printed shroud I think.

Here’s the thermal “calculations” with faked T_a to makeup for whatever intra-package resistance and thermal headroom baked into dynatron’s max TDP rating there might be:

A caveat to all this: I’m making the assumption that whatever testing dynatron did, it rolled up the die to ihs resistance into it which might not be a safe assumption.

On the silverstone cooler, I’ve seen the swap done and it was straight forward. For the coolserver, I have not seen it done but since it uses wire clips it shouldn’t be that difficult even if the stock fans are 28mm thick and the one you are replacing it with are 30mm thick (although I’m not sure how the middle fan swap would fare).

Its a combination of the heatsink+fan size, also the T30 is a really good/balanced fan. One of the reasons the dynatron j12 noctua fan swap isn’t great is that noctua just doesn’t make a very good 80mm fans; their high end 120mm and 140mm fans are decent though, if a bit too expensive.

These coolers will fit in a define 7 xl, it is a really big case.
Here’s a random ebay link to the cooler, I don’t necessarily endorse this seller:

Hey, I noticed it lists a whole array of compatible sockets:

Including BOTH sp5 AND sp6.

SP6 is supposed to be smaller than SP5. Does the cooler come with some sort of adapter? Or is it smaller footprint (SP6 is smaller than SP5) and, therefore, does NOT cover SP5 in its entirety?

Thanks!