Look ma, no radiator fans

[BHJohnson]

I finished a proof of concept for using a water cooling loop that cools the GPU and CPU, sealing up the case with the exception of the intakes for the fans and the exhaust for the top mounted 420 mm radiator, and turning off the radiator fans to try lowering overall noise signature. The result is that it works. My sealing job was fairly rough, I took a lot of short cuts with electric tape instead of actual gaskets, there are no gaskets on the fans whatsoever, I only turned the radiator fans off but didn't remove them, there's still some fairly good sized holes that I just ignored because I'm impatient, so there's a lot of both cosmetic and functional things that can be better about the idea, but it works. 20 minutes into a stress test and I'm sitting at a steady 70 C on the CPU, which is the hottest part of this system by a long shot. With a GPU and CPU stress test going simultaneously, temps on the CPU were about 77 C, which is warmer than with fans definitely, but again this was a quick and dirty test just to proof of concept the idea. Turning off the GPU stress test but leaving prime95 running, temperatures on the CPU came back down, proving that the system is removing heat and it's not just the thermal capacity of the coolant absorbing the heat input. I have no idea if I will tweak this idea in or what I will do with this information, but I feel satisfied having proven it can be done.

 

Side note, I really wish this board came with VRM temperature sensors.

[GoldenLag]

you shure you reached thermal equalibrium?

[mxk]

Cool. Sounds like the large mass of the radiator is helping to hold in heat so I think it might get hotter

[Tristerin]

15 minutes ago, GoldenLag said:

you shure you reached thermal equalibrium?

This - it takes longer than 20 minutes

[AngryBeaver]

49 minutes ago, BHJohnson said:

I finished a proof of concept for using a water cooling loop that cools the GPU and CPU, sealing up the case with the exception of the intakes for the fans and the exhaust for the top mounted 420 mm radiator, and turning off the radiator fans to try lowering overall noise signature. The result is that it works. My sealing job was fairly rough, I took a lot of short cuts with electric tape instead of actual gaskets, there are no gaskets on the fans whatsoever, I only turned the radiator fans off but didn't remove them, there's still some fairly good sized holes that I just ignored because I'm impatient, so there's a lot of both cosmetic and functional things that can be better about the idea, but it works. 20 minutes into a stress test and I'm sitting at a steady 70 C on the CPU, which is the hottest part of this system by a long shot. With a GPU and CPU stress test going simultaneously, temps on the CPU were about 77 C, which is warmer than with fans definitely, but again this was a quick and dirty test just to proof of concept the idea. Turning off the GPU stress test but leaving prime95 running, temperatures on the CPU came back down, proving that the system is removing heat and it's not just the thermal capacity of the coolant absorbing the heat input. I have no idea if I will tweak this idea in or what I will do with this information, but I feel satisfied having proven it can be done.

 

Side note, I really wish this board came with VRM temperature sensors.

You do know that you can get some 800 rpm fans that are basically inaudible right? Meaning you could increase the cooling capacity on the rad a good bit while also keeping the system completely quiet.

 

Here are some common items compared to sound levels. Keep in mind the average ambient noise level in a home can be around 8-12 DB sometimes more.

 

dBA	Example	Home & Yard Appliances	Workshop & Construction
0	healthy hearing threshold
10	a pin dropping
20	rustling leaves
30	whisper
40	babbling brook	computer
50	light traffic	refrigerator
60	conversational speech	air conditioner
70	shower	dishwasher
75	toilet flushing	vacuum cleaner
80	alarm clock	garbage disposal

 

A list of a few good options.

	Name	Price	Airflow (CFM)	Noise (dBA)	Max speed (RPM)
	Noctua NF-S12B REDUX 700	£12.96	33.5	6.8	700
	Scythe Slip Stream 500	£9.18	24.5	7.5	500
	Coolink SWiF2-120	£10.00	35.5	8.5	800
	Noctua NF-S12A ULN	£17.45	43.7	8.6	800
	Noctua NF-A12x25 ULN	£26.44	32.8	12.1	1200
	Nanoxia DS 120mm Fan	£9.29	60.1	14.2	1300
	Scythe Kaze Flex 120mm 800	£9.47	43.0	14.5	800
	be quiet SW3 120mm 3pin BL064	£18.37	50.5	16.4	1450
	be quiet SW3 120mm PWM BL066	£18.42	50.5	16.4	1450

[Mira Yurizaki]

I'm also slightly skeptical a radiator is effective for passive cooling because in order for passive cooling to be effective, the radiating fins have to be spread apart further. If they're close together as in most forced convection coolers, it's harder for air to naturally flow through it.

[AngryBeaver]

6 minutes ago, M.Yurizaki said:

I'm also slightly skeptical a radiator is effective for passive cooling because in order for passive cooling to be effective, the radiating fins have to be spread apart further. If they're close together as in most forced convection coolers, it's harder for air to naturally flow through it.

Agreed. The only time I have seen a fanless rad setup that actually did a decent job it was geothermal and they were burying the radiator in dirt to disperse the heat.

[BHJohnson]

It's not actually passive cooling, there's still fans pressurizing the case to push air out the radiator. I contemplated trying passive cooling, but you either need a different heat sink (the Earth or similar), or you need a wicked high delta T to create enough driving head to get enough air to flow, making passive cooling impractical or ineffective. Possibly for a very low power system like a stock clocked APU that you're OK with operating at very high temperatures, but even then I would have doubts. I won't be trying it.

 

And reaching thermal equilibrium was part of the reason I turned on the GPU stress test in addition to the CPU stress test to warm the loop up faster. Turning off the second stress test caused temperature to go down, which means more thermal energy was leaving than going in. The CPU stress test ran for another two hours and held steady, so yes, I'm going to call that equilibrium given it's only a 140 mL reservoir and doesn't have that much mass to heat up.

 

I acknowledge this isn't really a practical idea, this is more just a neat proof of concept that I'm actually fairly surprised worked.