Project Jalapeno

[crmaris]

It can be also used as a DC load but its purpose its to heavily stress CPU coolers.

Attached you will find a photo of what a colleague of mine (Seafalco, a fellow reviewer at TheLab.gr) has made for me.

The transistors will not be used in the beginning. Instead we will use 6 TEH100M2R50FE (http://gr.mouser.com/Search/ProductDetail.aspx?R=TEH100M2R50FEvirtualkey588100 00virtualkey588-TEH100M2R50FE).

Total estimate max Watts of the 6-resistors equals to 346W, enough to stress every mega-cooler out there.

In the start I plan to keep things simple (and cheap), so I will use 6 switches to toggle on/off the resistors and one switch for the PSU. Also with the use of a shunt I will measure the current drawn by the resistors and along with the 12V rail I will be able to measure the total watts. Temperature readings will be taken in the center of the plate through a 1mm diameter hole, were the probe will be placed (the hole is 5mm deep).

Later (when I will have more money to spent :D ) the switches will be thrown away and in their place I will put some usb controlled relays. Also, with the help of a Labjack and the appropriate module (op-ap amplifier) I will be able to take, in the same time, both current and volt readings. Of course a custom program will be made to control all above and monitor in real time all variables :)

The work log of a similar (but much more sophisticated project) can be found here (http://www.thelab.gr/air-cooling/inferno_cpu-heatsink-tester-79045.html). The man behind this project (Seafalco) is the one that made the plate for my project.

[crmaris]

some more pics as the project is running to its end.

The development of the control program has already started!

[Travis]

The surface contact should simulate a real CPU's IHS.

[crmaris]

It's very hard to simulate a real Cpu die. With this tool we will be able to evaluate the performance of heatsinks easily but we we will not be able to say, exactly, how much watts from a real CPU each heatsink could handle.

[allikat]

Any way you could de-lid an old CPU to provide the contact surface? On old P4 or AMD chip should suffice for that?

[shovenose]

Any way you could de-lid an old CPU to provide the contact surface? On old P4 or AMD chip should suffice for that?

and the point would be?

[crmaris]

The project is finished. More photos and the full story here (http://www.thelab.gr/apo-to-ergastiri/codename-project-jalapeno-88844.html).

Besides heatsinks it can be used as a small PSU tester too :)

[rafal_iB_PL]

You have some mad skills. Mechanic, electronic, programming...

[crmaris]

thanks :) But most of the mechanical staff in this project have been created by a fellow reviewer at TheLab.gr, Seafalco. I only modified the PSU case, the modular PCB and finally assembled all pieces together.

Seafalco is one of the best mechanics/electronics I have ever met.

[Travis]

A heatsink tester has a lot to do with mechanics, since the retention module greatly affects performance.

btw: watch out the conductive Drain pad under the MOSFET.

[crmaris]

Jalapeno in action! A teaser from Xigmatek's new cooler (Aegir) testing. Currently the load is 19.1A*12.15V=232.065W

http://www.youtube.com/watch?v=Da0ORnTsdY0&feature=player_embedded

[Tator Tot]

Just a heads up; but this sort of thing has been done on English sites (FrostyTech) and almost every time the results prove to be useless as they don't reflect a real world scenario.

Testing of any cooling device is really hard. Though I may suggest reading over how Skinnee Labs (specifically Cameron/Skinnee) does it (http://skinneelabs.com/thermalright-venomous-x/2/).
It's not very complicated, it's just time consuming.

Still; it's your ballpark. Just a bit of my opinion.
To be fair, I only recommend the way Skinnee Labs does it; as their results always seem to be the most accurate and closest to home as far as heatsink testing goes. (They're also the JonnyGuru of anything Watercooling.)

[crmaris]

First of all thanks for the heads up!

Unfortunately I will disagree with you. I think that if you have a steady platform that outputs exactly the same amount of heat (Watts) and you test in about the same conditions (although I use delta in my tests) then the result is accurate and the performance of each cooler is equivalent to the one it has in real life. Also frosty uses very few Watts (max. is 150) so it is natural that their results are not so accurate, since they not stress coolers to their max.

Also in our team we are lucky to have a true master of aircooling that he proves everything he does with mathematic equations and, lucky for me, this is the fellow that helped me built Jalapeno. The latter is fair simple to what he is preparing right now (with my software support).

Take a look if you want at this thread and although in Greek by the pictures you will understand a lot (since its a long thead you could start reading from the last 2-3 pages).

http://www.thelab.gr/air-cooling/inferno_cpu-heatsink-tester-79045.html

[Tator Tot]

You should see about doing side by side runs with a Core i7 or Phenom II x6 and see what kind of results you get.

I still feel it's going to inaccurately represent cooling dynamics on actual systems.

I know for a fact that the heatsink companies (Thermalright, Promilatech, Cooler Master, & Thermaltake) test their heatsinks on actual systems. While they may do computer simulations and mock-ups. The down-right testing comes to a real system that they're pushing with a variety of programs.

It only leads me to believe; that if your method was better, they would be using it.

Though; as far as I know, Thermaltake is the only company to say how many watts of heat their cooler can effectively remove. (220w on the Frio, and 240w on the Frio OCK)

Most just list something like "supports 140w TDP CPU's"

[crmaris]

I think its fairly simple. You have a constant (this is Jalapeno's output power) and a variable (the heatsink). So every heatsink is tested under exactly the same conditions, thus the error margin is to small.

The only problem is that we cannot accurately calculate the ration between CPU watts and Jalapeno ones. But this was not our purpose, since we only want to see which heatsink is the best under our testbed. Of course, in a real system other conditions may apply, as fan's orientation inside the case, ambient etc.

So far our results are in par with real life ones (on the contrary frosty in its charts shows Noctua NH-C14 to be better than NH-D14. This is because they test with very few watts. After 180 watts NH-C14 shows a significant increase in temps while D14 holds very good. )

As for the cooler companies I do not think that they design their coolers without the help of a similar machine. They may test in the end in real systems but the designing process should be done with the use of a loader. Actually I may ask a PR I know to clarify this :)

[McSteel]

Every heatsink has a heat capacity, and heat resistivity (or you could take it's reciprocal, heat conductance). As long as you're inside certain parameters (temperature gradient and total heat output) the temperature increases linearly with heat output, and decreases linearly with surface area. Once you saturate the heatsink, this linearity is lost and the temperature rise turns into an exponential function. A few points I'd like to ask about:

- What have you chosen as a definite, final method for mounting the heatsink?
- Do you only test in steady-state (meaning after reaching thermal equilibrium)?
- If you will do dynamic testing, what will you use as your heat gradient value?
- What kind of thermal paste will you use? What if a heatsink doesn't come with one?
- Will you test on open air only, or will you have a case setup with limited supply of fresh air?
- Will you be calculating and publishing the qualitative descriptions i.e. the limit of linearity (and thus specific heat capacity) and the approximate conductivity (W/m*K)?
- Are you able to maintain a constant room temperature for the tests? If not, will you compensate for the differences in room temperature?

Sorry if I'm swamping you with these questions, but I'm curious about your method, I'd like to see some proper reviews, and your PSU reviews are flawless in my opinion.

[crmaris]

You are very right in what you say in the first paragraph. Seafalco (the fellow reviewer that helped me in Jalapeno) has also given all relevant equations about this matters in Jalapeno's thread at TheLab (and in Inferno's one, the successor of Jalapeno).


1) If it is possible we install the heatsink using its default mounting mechanism. If not we use a custom one. In both cases the heatsink is very firmly installed on Jalapeno's base.

2) We test for 10 minutes. In all tests so far thermal equilibrium is achieved in less minutes but we continue till 10 minutes. In Inferno we will have a function that will calculate when equilibrium is reached so we wont have to wait so long for every test.

3) I didn't quite understand this q. What do you mean with the word dynamic? To change rapidly the load?

4) We use each heatsink's thermalpaste, if the heatsink does not come with any we use Noctua's one (because I have plenty of it). Here I would like to add that before we start testing we run a full circle of tests in order to allow the break in (I don't know if this is the right work in english) of thermalpaste. We have noticed significant differences in results.

5) With Jalapeno we test in open air, Inferno will have its own fully controlled environment.

6) This will be done with Inferno, since in Jalapeno we haven't installed much sensors and we do not collect data via software (we only control it). With Inferno we will collect temp. data from 6-8 probes and all calculations of various equations will be done in real time, so it would be relative easy to give this info (which will be valuable to experienced users).

7) In the tests room I have controlled (in some degree of course) temperature and we also measure the delta between Jalapeno's base temperature and ambient. Seafalco using some theory proved that delta is the safest and more accurate way.

I hope I covered your questions. I will ask Seafalco if he could assist me since this is his area of expertise (to tell you the truth I prefer PSUs :) ) and he knows much more than I do

[McSteel]

3) I didn't quite understand this q. What do you mean with the word dynamic? To change rapidly the load?

I meant doing a thermal hysteresis test, then a dynamic load pattern to determine it's influence. Namely, if you switch the Jalapeno (or Inferno) from off (ambient temp) to full on (say, 200W) there will be a lag in the HS warming with respect to tester plate warming. Then, from full-on to full-off there will be a lag in tester cooldown with respect to HS cooldown. This will result in a difference between simple math and real-world conditions in dynamic load situations. To clarify further, if your full-off (or idle, doesn't make too much of a difference) temp is 30° and your full-on temp is 70°, you'd expect your temperature in rapid on-off cycles to be around 50° (i.e. the mean value between on and off), but it will be closer to one of the extreme values (either idle or full load), depending on the HS heat capacity, HS aerodynamics and thermal paste properties; as well as on/off cycle duration. It would be nice to have temperature rise and fall curves (off-to-full and full-to-off), and a dynamic load curve... This is something you could probably obtain during testing, along the way, without dedicating separate time to it.

Thanks for answering all that :)
Going by those replies, this looks very promising indeed... You just might grow to like HS tests as much as PSU tests, in time ;)

[Seafalco]

Hi to you all!
First of all I must ask for your understanding , and I hope my English not to be impolite dew to mine lack of knowledge!
I feel obliged to McSteel for his questions ant propositions .

In Inferno project many of them are implemented, and dynamic tests for thermal “inertia” estimation.

I read your post and I’m thinking that here is something that deserve some more discuss:


Every heatsink has a heat capacity, and heat resistivity (or you could take it's reciprocal, heat conductance). As long as you're inside certain parameters (temperature gradient and total heat output) the temperature increases linearly with heat output, and decreases linearly with surface area. Once you saturate the heatsink, this linearity is lost and the temperature rise turns into an exponential function.

If we see a heatsink Thermal Performance Graph (i.e.)

http://www.aavidthermalloy.com/technical/graph.shtml

(Sorry for the inconvenience)

As you can see for the second triad of Watts (3-6 ) we have approximately 70 C temperature rising , bat for the third triad (6-9 Watts ) the corresponding temp rise is only 60 C.

The same think you will find it in the following article :

http://www.aavidthermalloy.com/technical/correct.shtml


Temperature Correction Considerations
Since natural convection heat sink efficiency degrades with decreasing sink-to-ambient temperature differential, a correction factor must be applied to the published data if an application requires a sink-to-ambient temperature rise of less than 75°C. The corrected thermal resistance is obtained by multiplying published °C/W/3-in data by the appropriate factor from the following table:
Temperature Rise (DTsa) Correction Factor
75°C 1.000
70°C 1.017
60°C 1.057
50°C 1.106
40°C 1.170
30°C 1.257

For any extrusion profile in natural convection,
the thermal resistance (°C/W) is more than 25% higher
at DTsa = 30°C than at DTsa = 75°C.
Length Correction Considerations

As you realize, when the temperature is higher the heat sink radiates more efficiently the heat, but the main issue here is that, to do it so the temperature rise might be “lethal” to heat producing device (i.e. CPU).


Of coarse when there is an forced air flow the things starts to differ, the curve is more strait because the initial curve is “stretched up” to a wider area of power .

Even so, hardly you can find something that you can call “Saturation” , the heat sink temperature will be higher and higher as the heating power increase , till to “melt down” of heating element .

There is a assumption: The behavior of the cooling procedure is “linear” , if we have a non linear heat sink we might have strong deviations from what we expect from a performance curve.

Specially for a heat-pipes based heat sink there is a point of strong deviation.
This is the point where the “cool” end of the heat pipe is hot enough to stop the phase change (steam to liquid) that suppose to happens there!
So the heat sink loose a crucial advantage , soon, there is no liquid into heating plate of the heat sink, the energy that before is consumed to evaporate the liquid (procedure that keeping the temperature “steady”), now is heating up some hollow copper tubes, who contains only steam!
This is the point that we measure an abrupt rising of temperature.

I hope my English not to be to “alien” to you, and we are in your disposal for any further questions.
Thanks again
Seafalco.

[McSteel]

That's exactly what I was getting at when I mentioned saturation. A uniform, homogeneous slab of aluminum is simple to describe and understand thermally. But when you throw heatpipes or vapor chambers into the mix, things get a bit more complex... Also, when you mix metals, internal resistance gradients between contact surfaces of different conductivity complicate things further... This is why only a thorough test can reveal the real limits of a heatsink, and where and how to best implement it.

Thank you for your reply, and I wish you the best of luck with your project! :)

BTW, don't worry about your English, it's more than adequate. And worlds apart from my Greek :p ;)