CPU Cooler Theory vs Reality

[ChalkChalkson]

Hey guys!

I recently exchanged a few comments with @AlexTheGreatish in the comment section on the floatplane video about the 10.000RPM fan.

We talked about modeling CPU coolers, and neither of us was quite sure how to predict heatpipes properly. In the end I found citations mentioning that they behave roughly like conductors and are in the range of 1.5-5 W/K for small heatpipes like in the typical tower cooler.

Now I want to collect data on it, especially the accuracy. I did the maths and the experiment for my CPU and cooler a few times and find that my cooler reaches about 85% of its theoretical limit.

So now I have made an excel table that does all the maths for you, als also gives a predicted temp, so that you can use it to estimate whether or not a cooler is enough for a given CPU.

 

I'd be extremely greatful, if you could fill in the cells with grey background for your coolers and cpus (AI suite gives CPU power draw, fan speed and temp in one place) and post the performance in % here in the thread, ideally for idle and load.

 

Mine are between 80-85% for load and 90-92% for idle.

 

Here is a google doc you can copy to your drive

 

 

EDIT:

-replaced excel download with google sheets link

-added thermal compound

 

[Enderman]

Heat pipes are actually more conductive than copper, which is 385W/mK

[LUUD18]

You should make it a google document so people can add their stuff without having to download something.

[ChalkChalkson]

2 hours ago, Jorgen297 said:

This is probably the hardest thing to calculate within computers because it relies on pure physics, and there's just an insane number of variables. I mean if Intel cheaps out on the thermal paste, even a high tier water cooler will be hot. 

Well, yeah... but really you have to look at the orders of magnitude, assume I ignore a step of separation which corresponds to, say 40 W/K, you then simply look at the delta T around that step (P/(40 W/K)) which is just around 1.25°C in my case, so the total error is off by around 2.5%. I might look up the thermal parameters of intel's thermal paste and add a cell for thermal paste. 

 

2 hours ago, LUUD18 said:

You should make it a google document so people can add their stuff without having to download something.

You already converted it, am I allowed to edit your link into my post?

 

2 hours ago, Enderman said:

Heat pipes are actually more conductive than copper, which is 385W/mK

Ok, let's estimate the conductivity of a 60mm long (average between the 120mm of a typical cooler height), 6mm diameter heatpipe that managed to deliver 2W/K... Putting it into wolfram alpha shows that this heatpipe would have a conductivity of roughly 4250 W/mK which is in deed way higher than copper

[ChalkChalkson]

Ok, I didn't find any good data on the internal compound used (especially since I'd need to know thickness and die sizes as well) but I added the external compound into the mix... I am using the 37.5mm * 37.5mm listed for the packed on arc intel for the consumer chips, I would like to add a column for the packet size, but that'd make the table ugly  

And the excel download was replaced with a google sheets link

 

With the new stuff added mine is up to about 95% under load.. I not sure whether this is a triumph for "be quiet!" (why is there punctuation in their brandname, the space and the exclamation mark look weird in a sentence?!) or for physics....

 

(BTW using this exact sheet with 2W/K for the heatpipes and data from be quiet coolers I now consistently reach about 80°C as predicted temps)

[AlexTheGreatish]

This is awesome.  I have access to heaps of coolers and can make sure the thermal compound variable is kept consistent.  The most difficult part I think is knowing what exactly the wattage of the CPU is since none of them actually output their TDP, that could be something else interesting to look into though.  Seems like a wicked start to a proper nergasum of a video haha

[ChalkChalkson]

21 hours ago, AlexTheGreatish said:

This is awesome.  I have access to heaps of coolers and can make sure the thermal compound variable is kept consistent.  The most difficult part I think is knowing what exactly the wattage of the CPU is since none of them actually output their TDP, that could be something else interesting to look into though.  Seems like a wicked start to a proper nergasum of a video haha

Asus AI suite doesn't show the current drawn unfortunately, but it shows the power draw... and basically all of that is converted to heat

[TheGamingBarrel]

21 hours ago, AlexTheGreatish said:

This is awesome.  I have access to heaps of coolers and can make sure the thermal compound variable is kept consistent.  The most difficult part I think is knowing what exactly the wattage of the CPU is since none of them actually output their TDP, that could be something else interesting to look into though.  Seems like a wicked start to a proper nergasum of a video haha

With a decent motherboard, you can read the voltage to the CPU Directly, not using software, you *should* be able to get a rough idea of power draw then.

[ChalkChalkson]

1 minute ago, TheGamingBarrel said:

With a decent motherboard, you can read the voltage to the CPU Directly, not using software, you *should* be able to get a rough idea of power draw then.

not really voltage gives you just one part of the equation, you also need to know the amount of current drawn, and that depends a lot on what the CPU is doing... even hitting the CPU with the same amount of different instructions will result in differences in power draw, so you either also need to measure the current on the VRMs directly (sucks to do) or rely on guessing the current, in which case the "power drawn" stat shown in software will probably be superior 

[ChalkChalkson]

Ok, update guys.

Was just working on designing my implementation of the DIY CPU cooler, when I realised, that this sheet does a terrible job of simulating designs without heat pipes.

 

Here why:

The model I am using in the google sheet, assumes constant temperature, thus making it all about total area and average heatpipe performance. When using a design without heatpipes the temperature gradient becomes pretty large, thus making the bottom area dissipate way more heat than the top one.

 

My solution

So I decided to quickly write up an excel sheet that uses finite part simulation of the CPU cooler (using 1mm steps, so the thermal paste isn't simulated too well). Issue being that there are of course infinitely many valid states for the cooler to be in, but only one equilibrium state. The excel sheet uses a 1:1 mapping of valid states of the cooler to a single parameter (I used the temperature of the base) and then optimises this parameter to find the equilibrium (based on total dissipation being equal to generated thermal power).

For finding this equilibrium I am just using the built in Excel Solver.

I know it would have been possible to implement it directly, but that would have been way more work and I really hate thermodynamics  

 

My excel sheet

CPU cooler I designed it for: project folder a rough sketch of the cooler

 

 

I am also working on a better solution for this issue:

On 31.5.2017 at 8:02 PM, AlexTheGreatish said:

The most difficult part I think is knowing what exactly the wattage of the CPU

I want to build an EPS extension that measures the drawn current in line. Currently waiting for some cheapo ampmeters from china.

If it works, and you are interested I might just invest the 10€ to get it to you  

[AlexTheGreatish]

9 minutes ago, ChalkChalkson said:

Ok, update guys.

Was just working on designing my implementation of the DIY CPU cooler, when I realised, that this sheet does a terrible job of simulating designs without heat pipes.

 

Here why:

The model I am using in the google sheet, assumes constant temperature, thus making it all about total area and average heatpipe performance. When using a design without heatpipes the temperature gradient becomes pretty large, thus making the bottom area dissipate way more heat than the top one.

 

My solution

So I decided to quickly write up an excel sheet that uses finite part simulation of the CPU cooler (using 1mm steps, so the thermal paste isn't simulated too well). Issue being that there are of course infinitely many valid states for the cooler to be in, but only one equilibrium state. The excel sheet uses a 1:1 mapping of valid states of the cooler to a single parameter (I used the temperature of the base) and then optimises this parameter to find the equilibrium (based on total dissipation being equal to generated thermal power).

For finding this equilibrium I am just using the built in Excel Solver.

I know it would have been possible to implement it directly, but that would have been way more work and I really hate thermodynamics  

 

My excel sheet

CPU cooler I designed it for: project folder a rough sketch of the cooler

 

 

I am also working on a better solution for this issue:

I want to build an EPS extension that measures the drawn current in line. Currently waiting for some cheapo ampmeters from china.

If it works, and you are interested I might just invest the 10€ to get it to you  

This really sounds like a job for MatLab, I might even have some code sitting around from a while ago that solves this but I'll need to look on my old hard drive to see.  Doing up fin equations in Excel is a real pain haha (many integrations by iteration later)

[ChalkChalkson]

On 24.7.2017 at 9:44 PM, AlexTheGreatish said:

This really sounds like a job for MatLab, I might even have some code sitting around from a while ago that solves this but I'll need to look on my old hard drive to see.  Doing up fin equations in Excel is a real pain haha (many integrations by iteration later)

^^I checked the results in Mathematica (what my uni uses) but I wanted something that I can put out here for people who don't have access to really expensive software to use.

If I find the time I'll write a little x86 or java app for this, if someone is interested

 

EDIT:

I thought I should add my Mathematica code:

cd := 237
w[h] := Piecewise[{{0.06,h<0.02},{0.1,h>=0.02}}]
d := 0.001
cv := 10.45
pow := 90
fins := 50
height := 0.14
sol = NDSolve[{D[p[h],h] == -cv*w[h]*t[h] , D[t[h],h] == -p[h]/(cd*w[h]*d) , p[0] == pow/fins , p[height] == 0} , {t,p} , {h,0,height}]
Plot[{Evaluate[p[h] /. sol], Evaluate[t[h] /. sol]}, {h, 0, height}]
Evaluate[t[0] /. sol]

where cd is the conductivity [W/mK], w(h) is the width at a given height [m], d is the thickness of the fins [m], cv is the convection [W/m²K], pow is the CPU power [W], fins is the fin count, height is the total cooler height [m]

Note that you still have to add the temperature you lose to the thermal paste, this can be done by making cd and fins a piecewise function as well. In general, you can make the cooler completely your own by just making all constants piecewise functions

[AlexTheGreatish]

2 minutes ago, ChalkChalkson said:

^^I checked the results in Mathematica (what my uni uses) but I wanted something that I can put out here for people who don't have access to really expensive software to use.

If I find the time I'll write a little x86 or java app for this, if someone is interested

There are a couple pretty good ones online already, like this one

[ChalkChalkson]

Just now, AlexTheGreatish said:

There are a couple pretty good ones online already, like this one

^^oops, ok, won't invest that time then

But tell me, you interested in the EPS amp meter? Would probably order the parts from china again (they take a while and are well... cheap)

[AlexTheGreatish]

7 minutes ago, ChalkChalkson said:

^^oops, ok, won't invest that time then

But tell me, you interested in the EPS amp meter? Would probably order the parts from china again (they take a while and are well... cheap)

For sure, if it works then I'm all for it

[ChalkChalkson]

Just now, AlexTheGreatish said:

For sure, if it works then I'm all for it

Ok, I'll update whether it works or not, if it works, I'll try to find your address somewhere  

[AlexTheGreatish]

4 hours ago, ChalkChalkson said:

Ok, I'll update whether it works or not, if it works, I'll try to find your address somewhere  

Yeah keep me updated!

[ChalkChalkson]

On 25.7.2017 at 2:59 AM, AlexTheGreatish said:

Yeah keep me updated!

Ok, back from holiday, and the cheapo amp-meters arrived. Unluckily only 3.

I quickly soldered it together, plugged powered it over LP04 and the lights turned on showing 0 as they should. So I took a 1.5V battery to it and shorted it over the EPS adapter. It should 3.XX A which my "proper" multi-meter confirmed, so the accuracy is within 1-5% .

I then put it in between the EPS plug on my PSU and the extension running to my CPU and turned the system on. 

Which I regretted within 1-2s because a cable caught fire.

When the fire was taken care of and I had time to look at the damage I realised something very funny: the cable that caught fire was the ground of my LP04 plug, which handled (and better should handle) the load easily when I did my accuracy test with the battery.

The only real solution I could imagine for that was, that the amp-meters are not built in a way, that allows the device power, and the measured current to share a physical ground anywhere (in this case the common ground sat inside the PSU, so I was VERY happy that I did this test with my RM850 and not some cheapo thing).

So I had to power them with a battery, luckily these things don't really care if you give them 5 or 9 volts, so I just ran them off of a 9V brick.

This time everything worked like a charm and the measured values I got when I plugged it into my PC actually made sense and told a buch of things about my setup I did not know, eg that there is a time during post, where my CPU draws a lot of current.

 

So yeah, the thing works as intended, too bad there are only 2 amp-meters left, but that actually isn't too bad, I measured on all channels, and the deviations are pretty small, so just adding the 2 amp values up and multiplying with 24 gives a pretty good estimate of actual power drawn.

 

Currently the little thing is piss ugly, an exposed cable mess with 2 displays on a 5x5 cm square of plastic and a heavy free floating battery case. So I might build a little enclosure. However and enclosure will make it more expensive to ship it to Cannada in case you are still interested. Without the case I will be able to mark it as a "Großbrief" which costs 3.7€, if I need to send it in a package it will be 16€  

I would happily "invest" both these sums, but only if I know whether you would actually try/use it. I know you guys' time is very limited and you receive a ton a stuff, so no one would really profit if I sent you something to rot on the new shelves. 

Thus it would be super great, if you could give be a bit of feedback which version you would like to receive if any at all.

 

Unfortunately I woun't be able to build a case in the near future, because I have a pretty important exam on Monday and then another one in the following week, so that version probably wouldn't leave my home until the 2. week of September at best

 

 

BTW:

There is no English datasheet for these little fckers, I could find one for identical ones, but that was written entirely in Kanji. There were instructions in the amazon descriptions, but those were obviously from google translate and even worse: were different for the English and German amazon pages, so I tried both versions, the English instructions lead to the fire, the German ones to my PC not turning on.... What ever the potential buyer may think: don't buy stuff without datasheets, it is dangerous

 

I also took some pictures, but this post is long enough as it is, so if anyone is interested I will just post them separately. Unfortunately none of the fire though, was a little busy at the time

 

Apologies for the abominable English are probably adequate, too. Just came back from sport and I am super tired  

See you tomorrow (for me),

 

Chalky

[AlexTheGreatish]

10 minutes ago, ChalkChalkson said:

Ok, back from holiday, and the cheapo amp-meters arrived. Unluckily only 3.

I quickly soldered it together, plugged powered it over LP04 and the lights turned on showing 0 as they should. So I took a 1.5V battery to it and shorted it over the EPS adapter. It should 3.XX A which my "proper" multi-meter confirmed, so the accuracy is within 1-5% .

I then put it in between the EPS plug on my PSU and the extension running to my CPU and turned the system on. 

Which I regretted within 1-2s because a cable caught fire.

When the fire was taken care of and I had time to look at the damage I realised something very funny: the cable that caught fire was the ground of my LP04 plug, which handled (and better should handle) the load easily when I did my accuracy test with the battery.

The only real solution I could imagine for that was, that the amp-meters are not built in a way, that allows the device power, and the measured current to share a physical ground anywhere (in this case the common ground sat inside the PSU, so I was VERY happy that I did this test with my RM850 and not some cheapo thing).

So I had to power them with a battery, luckily these things don't really care if you give them 5 or 9 volts, so I just ran them off of a 9V brick.

This time everything worked like a charm and the measured values I got when I plugged it into my PC actually made sense and told a buch of things about my setup I did not know, eg that there is a time during post, where my CPU draws a lot of current.

 

So yeah, the thing works as intended, too bad there are only 2 amp-meters left, but that actually isn't too bad, I measured on all channels, and the deviations are pretty small, so just adding the 2 amp values up and multiplying with 24 gives a pretty good estimate of actual power drawn.

 

Currently the little thing is piss ugly, an exposed cable mess with 2 displays on a 5x5 cm square of plastic and a heavy free floating battery case. So I might build a little enclosure. However and enclosure will make it more expensive to ship it to Cannada in case you are still interested. Without the case I will be able to mark it as a "Großbrief" which costs 3.7€, if I need to send it in a package it will be 16€  

I would happily "invest" both these sums, but only if I know whether you would actually try/use it. I know you guys' time is very limited and you receive a ton a stuff, so no one would really profit if I sent you something to rot on the new shelves. 

Thus it would be super great, if you could give be a bit of feedback which version you would like to receive if any at all.

 

Unfortunately I woun't be able to build a case in the near future, because I have a pretty important exam on Monday and then another one in the following week, so that version probably wouldn't leave my home until the 2. week of September at best

 

 

BTW:

There is no English datasheet for these little fckers, I could find one for identical ones, but that was written entirely in Kanji. There were instructions in the amazon descriptions, but those were obviously from google translate and even worse: were different for the English and German amazon pages, so I tried both versions, the English instructions lead to the fire, the German ones to my PC not turning on.... What ever the potential buyer may think: don't buy stuff without datasheets, it is dangerous

 

I also took some pictures, but this post is long enough as it is, so if anyone is interested I will just post them separately. Unfortunately none of the fire though, was a little busy at the time

 

Apologies for the abominable English are probably adequate, too. Just came back from sport and I am super tired  

See you tomorrow (for me),

 

Chalky

Hahaha holy crap that sounds like quite the adventure.  I was looking into how to properly measure the current and voltage on different parts of the motherboard a little while ago and concluded it was a lot easier to short something out than I originally anticipated, very different beast than the circuits I've dealt with before where you can basically prod where ever and be fine.

 

If you want to send a link of where I can buy one of them we can probably just purchase it ourselves since the shipping is pretty crazy and we might be able to get it for a fair bit less.  We also have an oscilloscope and a multimeter now so we probably can get the required information using those if you could include pictures of where to take the measurements.

 

 

[ChalkChalkson]

2 minutes ago, AlexTheGreatish said:

Hahaha holy crap that sounds like quite the adventure.  I was looking into how to properly measure the current and voltage on different parts of the motherboard a little while ago and concluded it was a lot easier to short something out than I originally anticipated, very different beast than the circuits I've dealt with before where you can basically prod where ever and be fine.

 

If you want to send a link of where I can buy one of them we can probably just purchase it ourselves since the shipping is pretty crazy and we might be able to get it for a fair bit less.  We also have an oscilloscope and a multimeter now so we probably can get the required information using those if you could include pictures of where to take the measurements.

My solution was a bit simpler: I just put the amp meters inline in an EPS connector, I got a ballpark of what is lost to the specific VRMs on my MoBo, by looking at their specsheet and went from there.

I will be able to check how close this comes to other popular measuring option like AI suite, when I am back at my flat, if the difference is less than 5% I would consider that good enough.

Measuring behind the VRMs is pretty darn crazy because of the very high current and low voltage. It might also introduce instabilities, because of the long lines and the voltage drop over the measuring resistor, so I would highly advise against it.

Another solution would be to use one of these ampmeters that have a ring on the front and measure the induced magnetic field instead of the current directly, but a PC is a pretty messy place when it comes to EM fields.

 

As of linking what I used: I just searched for "ampere meter" on amazon and bought the cheapest and smallest one that I could easily solder on an EPS extension 

[ChalkChalkson]

On 26.8.2017 at 12:36 AM, AlexTheGreatish said:

where to take the measurements.

I finally got an answer to that. It seems like people like der8auer take the measurement using an induction ampmeter like this one and measure around the 12V EPS connector.

This is a bit easier than my ugly little inline adapter, but a little more vulnerable to EM noise.