Comprehensive CPU cooling experiment

[OrdinaryPhil]

I have decided to perform the ultimate cooling experiment to decide which is better - AIO, or Air cooling.

Here is what I have so far.

Dell Optiplex 7010:   i7-3770

Dell Optiplex 7020:   i7-4770

Dell Optiplex 7040:   i7-6700
Dell Optiplex 7050:   i7-7700

Dell Optiplec 7060:   i7-8700

Dell Optiplex 7070:   i7-9700

 

All of these are what Dell refers to as 'mid tower' devices, which are fairly compact, but still too large to be considered SSF (small form factor).

I've chosen this hardware because - while it isn't perfect - it does normalize for a few things. I won't be using boards from different OEMs like ASUS, Gigabiyte, or MSI, and I won't be using PSU's from those vendors either. In each of these instances, I will be using the stock Dell PSU and board, and the only thing I will be changing out is the cooling solution. It also normalizes for air flow inside the case.

I will run the Intel XTU stress test on each device for an hour, to see how the cooling solution keeps up.

Air:     BeQuiet! Dark Rock Slim
AIO:   Cooler Master ML120L
 

 

So the results were about in line with my expectations, but the data was uncomfortably tight. I'll probably have to log the numbers again and record with an additional decimal so that the spacing is more legible. 

 

What's important to remember for this particular graph is that the temps were recorded while TurboBoost was disabled. Meaning the CPUs were all running at their base clock speeds. It will take me a while to run all these tests again, but I'll do so with TurboBoost enabled. It should be a pretty significant bump in temps.

This data paints in interesting picture;

Firstly, water cooling results in very similar similar temps to air at idle - although some were a tiny bit lower. The time it took for the temps to rise was longer with water - the CPU remained cooler for longer during the first 5 to 10 minutes, but eventually leveled off. After a full hour of stress testing, the air cooler was able to keep a mildly lower temp than water was. This is due to the fact that water takes longer to reach thermal soak than air, and there are more layers of abstraction between the AIO and the CPU than there are between the air cooler and the CPU - sort of.

Air coolers reached their maximum temp after abut 10 miuntes, where the water cooling didn't reach their maximum temps until somewhere between 30 and 60 minutes - even if the temps were a degree to two warmer. This tells me that for short, bursty workloads, water may be preferable, but for any work longer than 30 or 60 minutes, the air will provide a lower absolute value. We'll see if this trend continues at higher clock speeds (Turbo ON).

I'll post the second set of data in a week or so once I've had time to process it.

 

 

[dean_5234]

17 minutes ago, OrdinaryPhil said:

I have decided to perform the ultimate cooling experiment to decide which is better - AIO, or Air cooling.

Here is what I have so far.

Dell Optiplex 7010:   i7-3770

Dell Optiplex 7020:   i7-4770

Dell Optiplex 7040:   i7-6700
Dell Optiplex 7050:   i7-7700

Dell Optiplec 7060:   i7-8700

Dell Optiplex 7070:   i7-9700

 

All of these are what Dell refers to as 'mid tower' devices, which are fairly compact, but still too large to be considered SSF (small form factor).

I've chosen this hardware because - while it isn't perfect - it does normalize for a few things. I won't be using boards from different OEMs like ASUS, Gigabiyte, or MSI, and I won't be using PSU's from those vendors either. In each of these instances, I will be using the stock Dell PSU and board, and the only thing I will be changing out is the cooling solution. It also normalizes for air flow inside the case.

I will run the Intel XTU stress test on each device for 2 hours, to see how the cooling solution keeps up.

I haven't yet chosen which cooler to use for the test, but they will need to be as 'like for like' as possible. Here is what I am looking at presently:

Air:     BeQuiet! Dark Rock Slim
AIO:   Cooler Master ML120L

Both the cooling solutions have a fin stack of approximately the same size, so they will have approximately the same thermal mass in copper/aluminum (not factoring in the liquid). Both have a single 120mm fan. Both are $64.99 USD on Amazon.

If there is a better match between the AIO and the Air Cooler I will happily hear suggestions.

Thoughts?

That's a cool experiment! Personally I'd be interested in the Corsair A500 air cooler as it is currently 45 bucks but looks as if it could outperform an AIO

[NorKris]

With the word ultimate you mean budget coolers and budget cpu's. hmm ok

[cummerou1]

21 hours ago, Anonymous Joe said:

That's a cool experiment! Personally I'd be interested in the Corsair A500 air cooler as it is currently 45 bucks but looks as if it could outperform an AIO

There's a reason why it is 45 bucks, it's baaaaaaaaad
 

 

[OrdinaryPhil]

5 minutes ago, cummerou1 said:

There's a reason why it is 45 bucks, it's baaaaaaaaad
 

 

We are trying to keep things as 'like for like' as possible.

Whatever Air cooler we settle one will need to be of approximately the same mass as the AIO, and will be limited to the same number of fans.

Because the ML120L is fairly small, it would be unreasonable to compare it to something like the A500 (however poorly it performs) simply on the grounds that A500 would be in a functionally different 'weight class'

[cummerou1]

1 minute ago, OrdinaryPhil said:

We are trying to keep things as 'like for like' as possible.

I didn't say it couldn't be used, just that it definitely will not outperform a normal AIO like Joe suggested it might

[OrdinaryPhil]

1 minute ago, cummerou1 said:

I didn't say it couldn't be used, just that it definitely will not outperform a normal AIO like Joe suggested it might

I still think it would be unfair to compare them.

Coolers that we think are good comparisons will be limited to a single 120mm fan. So things like the Dark Rock Slim, or NH-U12s are going to be more reasonable fits. Even something cheaper like the Hyper 212 Black is going to be a better comparison.

[cummerou1]

1 minute ago, OrdinaryPhil said:

I still think it would be unfair to compare them.

Coolers that we think are good comparisons will be limited to a single 120mm fan. So things like the Dark Rock Slim, or NH-U12s are going to be more reasonable fits. Even something cheaper like the Hyper 212 Black is going to be a better comparison.

Sorry, I only saw your top line ( "we're trying to keep things like for like")

Again, I am not encouraging you to use it, not at all, I just wanted to tell Joe why it is that the A500 does not outperform an AIO, and why it is so cheap compared to other air coolers.

I definitely agree that you should actually use a "proper" air cooler

[LogicalDrm]

Maybe calling this anything "ultimate" is bit reaching. You are testing budget upgrade for OEM build. In general reviewers run coolers in open test-benches to eliminate case as a factor in results. So there test is per cooler instead of per system. There are some who use normal cases with normal fan layout (2 in, 1 out for air, 2 in, 2 out for AIO). For you the case design of Dell will be a factor. And can be important as such since those cases usually lack in airflow.

 

There's also other notes, like how you conduct fan controls or do you let them go auto. And the lack of pushing the limits since there won't be any OCing.

[OrdinaryPhil]

1 hour ago, LogicalDrm said:

Maybe calling this anything "ultimate" is bit reaching. You are testing budget upgrade for OEM build. In general reviewers run coolers in open test-benches to eliminate case as a factor in results. So there test is per cooler instead of per system. There are some who use normal cases with normal fan layout (2 in, 1 out for air, 2 in, 2 out for AIO). For you the case design of Dell will be a factor. And can be important as such since those cases usually lack in airflow.

 

There's also other notes, like how you conduct fan controls or do you let them go auto. And the lack of pushing the limits since there won't be any OCing.

Gotta disagree. My English was on point - it's the ultimate experiment - not experimenting with the ultimate parts.

The HEDT / enthusiast market is very tiny and I'm not concerned with collecting data that only applied to 0.5% of the PC market. This experiment is intended to be practical and informative to a large group of users that often use older hardware and want to know how much to manage/maintain it.

But thanks for being the language police, lol

[OrdinaryPhil]

1 hour ago, LogicalDrm said:

There's also other notes, like how you conduct fan controls or do you let them go auto. And the lack of pushing the limits since there won't be any OCing.

This is a good thing to discuss.

I'm probably going to run a LOT of tests, but I don't want to end up with a graph or chart that is so packed that it becomes illegible. I'd like to run tests with each CPU under different configs.

 

The end result is going to be a graph charting the temps of each CPU over time. I'll have a graph for them running at their base clock (turbo off), and a graph for them running at their max turbo frequency. I won't be doing anything with OC since that's so niche, and since each CPU handles it so differently that the data wouldn't actually be representative.

I may try to get my hands on the 'k' series of each CPU in the future and do the same experiment with those since they will have slightly higher base/turbo clocks.

Those are really the only things like that I am concerned with graphing. The fan curve will be set to default. The RAM we are using is also at base speed, no OC on the RAM to normalize for that as well. That means using DDR3 1600 for the DDR3 compatible CPUs, and DDR4 2666 in the DDR 4 systems. In both cases the arrangement with be 4x4GB from Samsung. CAS and other timings will be set to default. XMP disabled.

The reason for this is that most people are not making tweaks in their BIOS, and these settings represent the defaults that most users will experience. Again, the "ultimate" experiment is designed to be as useful and practical as possible, and to represent a healthy chunk of the PC demographic, not just the extreme few. Those people are not going to benefit from this info anyway.

 

[LogicalDrm]

20 minutes ago, OrdinaryPhil said:

The end result is going to be a graph charting the temps of each CPU over time. I'll have a graph for them running at their base clock (turbo off), and a graph for them running at their max turbo frequency. I won't be doing anything with OC since that's so niche, and since each CPU handles it so differently that the data wouldn't actually be representative.

You call OCing here niche, as well as stating HEDT/enthusiast being niche overall. Yet your sample is going to be very niche by itself. Dell may be the most used OEM in US, but thats about it. Testing in Dell cases is going to be niche too since results are tied to how case handles airflow.

 

For OCing, if you are using Optilex hardware, the mobos and BIOS's could also have some difficulties. Same would go to fan controls, but luckily those can be bypassed by couple of ways.

 

43 minutes ago, OrdinaryPhil said:

But thanks for being the language police, lol

Considering English is not my main language, I had to check the meaning. And still have to disagree use of the term here. The target of this test isn't something that could be applied just like that to any situation. Of which would make it ultimate or final test every to be needed. For me "ultimate" doesn't automatically refer to top end hardware, but to "one-which-covers-all" type of thing.

[Mark Kaine]

1 hour ago, LogicalDrm said:

For me "ultimate" doesn't automatically refer to top end hardware, but to "one-which-covers-all" type of thing.

That's true,  OP could throw in one or two modern "front mesh" cases with better airflow, then "ultimate" would really cover a wide range,  but I still think this is a cool experiment. 

 

And having bad airflow cases by itself should be actually interesting since it'll test the cooling capabilities of those coolers to the max! 

 

It's just in a test like this you still need fact check under different conditions,  so being limited to one type of setup isn't ideal. 

[OrdinaryPhil]

I've updated the original post with the data from the first series of tests