Experimentally determining the coldest I can get with air

[JamesWatt1]

Summary
I’m trying to see how cool I can get my PC using air cooling. The ideal configuration so far has been replacing the stock heat sink with a Hyper 212 and using the rear case fan only. I am in the process of testing a duct idea. I am considering ways to cool the air before it enters the case and upgrading the fans.

Details
I had this PC:

Intel Core i5-2500K
Stock heatsink/fan
SAPPHIRE Radeon HD 6950 2GB
G.SKILL Ripjaws Series 4GB DDR3 1600 9-9-9-24
Crucial RealSSD C300 64GB SATA III
ASRock P67 PRO3
COOLER MASTER HAF 912
XFX P1-650X-CAG9 650W ATX12V 2.2
Microsoft Windows 7 Home Premium 64-bit

Before I switch over to water, I want to see what I can do with air cooling.

Here are the changes I’ve made so far and what I’ve gotten from them. To measure the effect of a change, I ran LinX, Prime 95, and SuperPi for 10 minutes. During the last minute, I took the average temperature measured using the software that came with my motherboard (ASRock Extreme Tuning Utility). Ambient temperature was measured by a thermometer on the desk above my computer.

1) I replaced the stock heat-sink and fan with a Cooler master Hyper 212 Plus with Arctic Silver 5.

In both cases, I let the bios control the clock (it chose 3.3 Ghz) and the voltage (it chose 1.216 Vcore).

60 degrees C: stock heat-sink (ambient 18 degrees C)
49 degrees C: Hyper 212 & AR5 (ambient 15.6 degrees C)

As an aside, one of the challenges I’ve been facing is maintaining a constant ambient temperature. In NJ, the weather has been really inconsistent and I’m too cheap to pay for heating. I’ve tried opening and closing my windows to various degrees, but it’s not very precise. Any ideas?

2) I experimented with various degrees of case enclosure.

In all cases, I set clock to 4.7 Ghz (47 * 100), voltage to 1.35 Vcore, short & long duration power limits to 500.

70 degrees C: default case (ambient 17.5 degrees C)
65 degrees C: open case—top, sides, and pci slot covers removed (ambient 17.5 degrees C)
72 degrees C: sealed case—I taped cardboard over all holes in the case without fans in them (ambient 18 degrees C)

3) I experimented with different case fan locations (I was inspired by this article http://tech.icrontic.com/articles/pc_airflow_heat_cooling_guide/)

In all cases, I set clock to 4.7 Ghz (47 * 100), voltage to 1.35 Vcore, short & long duration power limits to 500. The case fans are Coolermaster A12025-12CB-3BN-F1 fans (120 MM, 1200 RPM, 44 CFM, 19 dB). I managed to keep ambient at 18.9 degrees C for all of these trials.

65 degrees C: one fan low in the front and one fan high out the back (default HAF 912 configuration)
60 degrees C: one fan high out the back
65 degrees C: one fan high in the front and one fan out high in the back
66 degrees C: one fan high out the back and one fan out the top of the case
66 degrees C: one fan out the top
67 degrees C: one fan low in the front
67 degrees C: one fan in from the side

4) A co-worker suggested that I try a duct. His argument was that doing so would allow me to maximize the rate that air flowed past the heat-sink (and consequently convection).

65 degrees C: one fan low in the front and one fan high out the back (ambient was 18.9 degrees C)
67 degrees C: duct as pictured below (ambient was 17.8 degrees C)

http://postimage.org/image/2akrrc3o/

I wasn’t thrilled with this result, but there were a couple of issues with this version of the duct.
-I couldn’t bend the duct from the heat-sink to the back of the case such that it lined up perfectly with the rear case fan hole (about half of the cross section is blocked). I’m thinking about taking a hole saw to the case, though would be open to a non-case damaging alternative.
-Also, the seals in the couplings that I machined weren’t perfect. I’m going to seal them with Loctite high temperature RTV.
-I’m not convinced that the ideal duct fan configuration is one on the heatsink and one in the rear. I’d like to try out a couple of different configurations to see what works best. If I had to guess, I would think based on my experience with fan configurations in the case that back only would be best.

5) Other things to try

-One idea I was kicking around was finding a way to cool the air before it is taken into the case. I was thinking of either something like an electrically cooled lunch box (like the Koolatron P20 http://www.google.com/products/cata...=X&ei=uMSLTd61HIi30QGH45ijCw&ved=0CD4Q8gIwAA#) or a mini-fridge. My concern with these is the huge amount of electricity needed to generate heat.
-Another direction is changing up the fans I have. My 120mm heat-sink fan only moves about 75 CFM. I bet I could find something that moves more air but that doesn’t sound like an air plane.

Which of these do you think is most promising? Can you think of a better idea?

 

[Zap]

Nice ducting work. Too bad it didn't work out. What you might need to do with the ducting is to double up on the fans because it may be resulting in lower CFM passing through the whole thing.

Actually, just doubling the fans on the 212+ would probably help a few degrees. Doesn't it come with an extra set of fan clips?

For much better results you will have to get a better heatsink. The 212+ is not popular because it is the best, because it isn't. It is popular because it is (often) the cheapest that use 120mm fans, ensuring a minimum level of performance and noise characteristics.

 

[Schmide]

One thing that will pop up when you eliminate/concentrate airflow away from the socket is heat creep from the VRM/Northbridge etc. This was first noticed by Watercoolers when thermal diodes were off the chip. Even thought they had superior cooling their idle temps were often higher than even mediocre air. Many, including myself, still make sure there is ample cooling/airflow around the socket despite any exotic cooling applied to the top of the chip.

 

[JamesWatt1]

Nice ducting work. Too bad it didn't work out. What you might need to do with the ducting is to double up on the fans because it may be resulting in lower CFM passing through the whole thing.

Thanks. What is your argument for doubling up on fans? Why would the CFM reduce as it goes through the ductwork? Would double fans increase turbulence?

Actually, just doubling the fans on the 212+ would probably help a few degrees. Doesn't it come with an extra set of fan clips?

Thanks for the idea. I picked up:

1x Rosewill RFA-120-BL 120mm 4 Blue LEDs LED Case Fan (it has CFM in the high 70s like the one that comes with the 212+)

2x SILVERSTONE FM121 120mm Case Fan. These turn in 110 CFM.

With the Rosewill, I wanted to see what I get from doubling up fans on the heatsink. With the silverstones, I'm going to find out what I get from increasing the CFM of the fans.

For much better results you will have to get a better heatsink. The 212+ is not popular because it is the best, because it isn't. It is popular because it is (often) the cheapest that use 120mm fans, ensuring a minimum level of performance and noise characteristics.

That's a good point. I added this to my list of things to try.

 

[JamesWatt1]

One thing that will pop up when you eliminate/concentrate airflow away from the socket is heat creep from the VRM/Northbridge etc. This was first noticed by Watercoolers when thermal diodes were off the chip. Even thought they had superior cooling their idle temps were often higher than even mediocre air. Many, including myself, still make sure there is ample cooling/airflow around the socket despite any exotic cooling applied to the top of the chip.

Thanks. I didn't consider this. When I'm fiddling with the duct, I'll put a fan in the space on top of the case.

 

[Zap]

Thanks. What is your argument for doubling up on fans? Why would the CFM reduce as it goes through the ductwork? Would double fans increase turbulence?

For doubling the fans, reviews always show lower temperatures.

Regarding the duct, the way axial fans move air is in a cone shape, not straight through.

 

[zagood]

To get your cpu as cold as possible with air, the best thing you can do is duct directly from an intake. Hard to do with most cases, but it's possible with some effort and a dremel. Usually what you see is an intake toward the top of the case where the optical drives normally are, and leaving another intake for the rest of the case on the bottom front or side door.

 

[aigomorla]

been messing with super cold air as well.

Ummm... realized... my sensors arent calibrated to go down that low.

Water coolant temp is higher then what my CPUID is reporting.

You'd even assume my 580GTX to be able to report proper values.
Well, it doesnt... it idles BELOW ambient.. which means either im breaking physics, or sensors are wrong.

So i am going with uncailibrated sensors.



OP, you should get a hard temp measurement, because i dont think sensors were meant to go down that low.

 

[JamesWatt1]

To get your cpu as cold as possible with air, the best thing you can do is duct directly from an intake. Hard to do with most cases, but it's possible with some effort and a dremel. Usually what you see is an intake toward the top of the case where the optical drives normally are

Could you elaborate on how this is different from what I have? I think the duct intake is where the optical drives are. Here is a picture with the grill on the front of my case taken off (the optical drive covers are plastic mesh when the front cover is on).

http://postimage.org/image/2x4ic8g3o/

and leaving another intake for the rest of the case on the bottom front or side door.

Am I understanding you correctly if you are suggesting a similar duct for cooling my GPU?

 

[JamesWatt1]

been messing with super cold air as well.

Those are some awesome temperatures that you're putting up. I have a ways to go before I hit single digits and I need do do my own temperature measurement. Could you elaborate here (or point me to a thread where you do) on your rig: How are you making the air super cold? How are you routing it to your CPU? How are you exhausting the air after the CPU has heated it??

 

[Blain]

Ducts are the cheapest, most efficient, yet ugliest way of drawing cool air into your HS fan.

 

[aigomorla]

Those are some awesome temperatures that you're putting up. I have a ways to go before I hit single digits and I need do do my own temperature measurement. Could you elaborate here (or point me to a thread where you do) on your rig: How are you making the air super cold? How are you routing it to your CPU? How are you exhausting the air after the CPU has heated it??

the computer pulls air from a section of the room which i have fans constantly running though it.

So at midnight times, the room can get down to like 15C.

But those are the temps im working with and noticing sensors not operating properly.



Also my system is quite massive. :T
It messures 36x48x36 inches.. its a rectangular box.
Theres a reason why i put Watercooling God Mod on my signature.

 

[wwswimming]

i wonder what cools better - high humidity air at 33 degrees F, or low humidity air at about 22 degrees F.

 

[Blain]

Air density aside, lower humidity air would decrease the possibility of component (part or contact), corrosion.

 

[God Mode]

If you can, try to get your hands on another 212+. IME, the QC on these is utter crap and 2 of them had a temp difference of 6ºc.

 

[Modelworks]

one thing you need to look at is not using CFM for fan performance. Fans only deliver that CFM with nothing in front or behind them in free air. Put anything in front or behind of them or change the enclosure they are installed in and that CFM goes way down.

Always look for fans that have the pressure rating listed. If it isn't listed then you can't know how the fan performs in real use.
graph of how pressure effects performance of a sample fan, each fan is different.

A really good read on fans that covers air pressure, heatsink placement etc, is here:
http://www.nidec.com/aircooling/fantech.htm

 

[wwswimming]

that looks familiar.

once i had to cool a 10 kilowatt radar tube (10% duty cycle) and all i had was a 4 inch air duct.

i used a 4" diameter fan that cost $5000, from EG&G Rotron.

it worked. it was loud, hovercraft quality.

they had just gotten in a new lead electrical engineer and he blew his stack. had me transferred to another department.

they ended up using the $5000 fan - no choice.

if you really want to cool using air, and you aren't conducting the test in Northern Dakota or Northern Siberia (you know, 50 below zero), high airflow is very helpful.

also little details like heat sink geometry - maximizing heat transfer while presenting the minimum impedance to air flow.

i think stuff like this is fun ... unless you're working on cooling the reactors & fuel pools in Japan.

 

[deimos3428]

60 degrees C: one fan high out the back
...
Can you think of a better idea?

If I'm reading the results right, the best result to date is just a single case fan exhausting out the back on high, so start from there. When you added either the front intake fan or top exhaust fan to the mix, the temperature went up ~5C. I suggest you try a simple experiment:

1. Start again with only the single rear exhaust on high, all other case fans off.
2. You didn't mention your 212 fan configuration; anything but completely passive is fine.
2. Let the system run for a bit, then note the temps as a control.
3. Cover the large roof vent completely with a sheet of paper or two. It should stick due to negative pressure.
4. Let the system run for a bit, then note the change in temperature, if any.

All else being equal, reducing the number of vents in your case forces the cool intake air to move faster through any remaining vents. Depending on the resulting change in airflow, you may end up with cool air going through the HSF faster, but if not -- well, at least it's very easy to undo.

Also, you've a rather powerful GPU there. I'm completely guessing, but adding any fan as intake at this point might kill whatever negative pressure you have. But you might try adding the side fan as *exhaust*. This will assist the main exhaust in generating negative pressure, and may pull heat out of the main air path, if it doesn't disturb it too much.

 

[JamesWatt1]

6) I experimented with different heat-sink fans and heat-sink fan locations.

Experiment
All of these tests were done with a high back exhaust case fan and no duct. In all cases, I set clock to 4.7 Ghz (47 * 100), voltage to 1.35 Vcore, short & long duration power limits to 500.

The Stock 212+ fan does 75 CFM
The Rosewill RFA-120-BL does 75 CFM
The SILVERSTONE FM121 does 110 CFM

Results:

64 degrees C: Stock 212+ fan front intake (15.6 C ambient)
63 degrees C: Stock 212+ fan front intake & Rosewill RFA-120-BL fan back exhaust (15.6 C ambient)
62 degrees C: Rosewill RFA-120-BL fan back exhaust (15.6 C ambient)
62 degrees C: SILVERSTONE FM121 front intake and back exhaust (14.8 C ambient)
63 degrees C: SILVERSTONE FM121 back exhaust (14.8 C ambient)
62 degrees C: SILVERSTONE FM121 front intake (15.1 C ambient)

Implications
Surprisingly, it doesn’t seem that increasing the CFM of the fan has much of an impact. Less surprising was the fact that the higher CFM fans were louder (FM121’s generate 45 dB at desk level vs. 35 dB with Hyper212+ stock fan as measured by the Sound Meter app on my Droid 2)

The case for exhaust fan only in all situations is getting stronger. In the thread I linked before and in my case fan location experiment, I found that rear fan only produced the coldest temperatures. I’m seeing a similar thing here. Both the 75 and 110 CFM fans mounted on the rear of the heatsink, directed to exhaust produced the coldest or near coldest runs.

Future actions

a) If I found that the rear fan is important here, then the partially blockage on the rear fan in the duct might explain its poor performance. I'm going to track down more flexible ducting to see if I can get a clear shot at the rear fan hole. While I have the duct set up, I'm going to try different fan configurations (e.g., front of duct only, rear of duct only, both), including the top, north-bridge cooling fan configuration recommended by zap.

b) Deimos3428, this is an interesting variation on my enclosure experiments (experiment 2 in my first post). I'll give it a try.

c) Modelworks: thanks. I didn't know that static pressure played a role in fan performance. Of the heat-sink and case fans that you've seen, what has been the range of static pressures that you've seen? It will be interesting to test two fans of the same cfm ratings but different partial pressures

 

[JamesWatt1]

that looks familiar.

once i had to cool a 10 kilowatt radar tube (10% duty cycle) and all i had was a 4 inch air duct.

i used a 4" diameter fan that cost $5000, from EG&G Rotron.

it worked. it was loud, hovercraft quality.

they had just gotten in a new lead electrical engineer and he blew his stack. had me transferred to another department.

they ended up using the $5000 fan - no choice.

if you really want to cool using air, and you aren't conducting the test in Northern Dakota or Northern Siberia (you know, 50 below zero), high airflow is very helpful.

also little details like heat sink geometry - maximizing heat transfer while presenting the minimum impedance to air flow.

i think stuff like this is fun ... unless you're working on cooling the reactors & fuel pools in Japan.

That sounds awesome. What cooling did you all have before you got that monster fan? How hot was it before and after?

 

[Morg.]

Well, for what little testing I have done, you might wanna try 200mm fans, they rock @ cooling and they make no noise at all (the ones from CM I used anyway).

For example, when replacing a 120mm 2400RPM fan with one of those + cardboard ducting (total fail test), I ended up 10 degrees lower. (200mm 900RPM fan)

With a bit of airflow control and some bigger fans I think you can probably go down to 55°C

 

[JamesWatt1]

Well, for what little testing I have done, you might wanna try 200mm fans, they rock @ cooling and they make no noise at all (the ones from CM I used anyway).

For example, when replacing a 120mm 2400RPM fan with one of those + cardboard ducting (total fail test), I ended up 10 degrees lower. (200mm 900RPM fan)

With a bit of airflow control and some bigger fans I think you can probably go down to 55°C

I would love to drop 10 degrees.

I'm curious about what you mean by "cardboard ducting (total fail test)" Do you have pictures/can you give more description?

 

[Morg.]

Myeah I posted pictures of that in a "ghetto mod" thread on anandtech ...



So that's a 200mm fan from coolermaster, a cardboard cereal box from Aldi, the heatsink of the Zerotherm FZ120, and then my OC'd 6750 @ 4Ghz /1.5V .. did like 61°C prime max with that thing, vs 70°C with the 2400RPM 120mm fan provided.

Also you may notice the motherboard is standing in the middle of the case, I think I rested the graphics board somewhere on a hdd or w/e .

Now, I did a bit more research after that and it would seem you can get more performance / noise out of good 120 / 140 mm fans "for the same surface", which leads me to think that a 240mmx240mm array of scythe gentle typhoons would probably beat the shit out of this cardboard setup, which was based on a barely smaller area reduction with a fan that has next to no static pressure.

Keep in mind you're going from 400 cm² to 144 cm² when you duct 200mm stuff on a 120mm heatsink and that means you lose a lot if you lack static pressure, which gentle typhoons have way more than this coolermaster thing (which is perfectly silent by the way).

All in all, think outside the box, drill fan holes in your side pannel and top pannel, put more fans, put better fans (for cases and silence, slipstream, for air pressure gentle typhoon or SL-Flex, read reviews on spcr), remove everything that is a wall and should be an air hole, your tests showed clearly the main issue with computer cases : too many walls, not enough holes.

And mainly, why do small when you can go big, why should your CPU be cooled by a ridiculously small 120mm or 140mm fan when there's so much more available out there. Add a fan blowing on the socket itself, why not ?

You can also have a front case fan that goes through an air pipe that ends on your graphics board or even shrouds it (without impeding the fan, of course ), etc.

 

[Zap]

So that's a 200mm fan from coolermaster, a cardboard cereal box from Aldi, the heatsink of the Zerotherm FZ120, and then my OC'd 6750 @ 4Ghz /1.5V .. did like 61°C prime max with that thing, vs 70°C with the 2400RPM 120mm fan provided.

LOL, that's hilarious!

You know, part of why it worked well was because of the fact that the fan wasn't right against the heatsink. When the fan is right against the heatsink, there's a "dead spot" with no airflow in the center. I seem to recall that some companies sold fan spacers for radiators, for the hardcore liquid cooling enthusiasts. See these fan shrouds. "Many reviews have proven that fan shrouds increase the cooling capacity of radiators."

 

[deimos3428]

Yeah..um...on that note, just think of your entire case as huge fan shroud. Note in the original post the temps went down by 5C with a single exhaust vs. open air.

For the best air cooling, you want to direct a clear path of cool air into the case, over/through your hot components, and back out again. You don't necessarily need huge fans to accomplish that, or many of them.