Jump to content

Is this even going to work…

ColinLTT

Not to leave our viewers dissatisfied, we built a furnace of a PC, a wind tunnel with 7 radiators, and our own data logging system to find out once and for all the EXACT effects of stacking radiators on water temperature.

 

Check out https://learn.adafruit.com/thermistor/using-a-thermistor for a good guide on how to use a thermistor-based sensor.

Arduino Code: https://create.arduino.cc/projecthub/user1590239/wind-tunnel-water-sensors-28cf12

 

 

Buy AMD Threadripper 3970X
On Amazon (PAID LINK): https://geni.us/Th48DG
On Newegg (PAID LINK): https://geni.us/bBIusr5

 

Buy Alphacool NexXoS ST30 Radiators 
On Amazon (PAID LINK): https://geni.us/Mhq2Qns
On Newegg (PAID LINK): https://geni.us/L3XfABJ

 

Alphacool Temperature Sensor 
On Amazon (PAID LINK): https://geni.us/UClFi
On Newegg (PAID LINK): https://geni.us/e9YA58q

 

Buy Nvidia Titan V
On Amazon (PAID LINK): https://geni.us/4tAnQF
On Newegg (PAID LINK): https://geni.us/DOGtb

 

Buy Noctua NF-F12 
On Amazon (PAID LINK): https://geni.us/nbwWJ
On Newegg (PAID LINK): https://geni.us/eJuzz

Link to comment
Share on other sites

Link to post
Share on other sites

I am glad they tested Cross vs Co current flow, because Cross current is the most common configuration used in Heat exchangers in big Chemical plants, because it keeps a delta throughout the exchanger image.png.b65752888cdbb6c9aef812e2bc4b07cd.png

 For the stacked the air pressure would drop over every exchanger reducing turbulence a bit and all the mechanical energy being put into air causes it to heat up, along with lower Delta T we get lower heat transfer by final radiator but with the single rad, a higher turbulence and a high delta T would make it perform as good as the stack

In a chemical plant reducing capital cost is imperative so minimal size with maximum surface area, turbulence and delta T with a minimum LTTD (lower terminal temperature difference) usually 5/10 celcius, in Cross current configuration,like in test 2,  are always a must, keeping vibration in check though 😅

 

PC: Alienware 15 R3  Cpu: 7700hq  GPu : 1070 OC   Display: 1080p IPS Gsync panel 60hz  Storage: 970 evo 250 gb / 970 evo plus 500gb

Audio: Sennheiser HD 6xx  DAC: Schiit Modi 3E Amp: Schiit Magni Heresy

Link to comment
Share on other sites

Link to post
Share on other sites

You know a more interesting way to test this would have been to use a rack mount configuration. Assuming the fans are 25mm and the rads are the "monster" thick 80mm ones - that's about 4.5 inches per set. Times 7 plus a pull fan at the back would be a stack of 7 rads plus fans roughly 800mm or 31.5 inches deep. Any standard rack can handle that depth and also facilitate exhausting that heat away out the back.

 

Do the same with more potent 38mm finger-chopping Deltas and you have a depth of roughly 900mm or 35.5 inches.

 

Any 3x120 rad will fit in a 17 inch wide rack profile and there are cover plates available for 3x120mm fans that do exactly this.

 

Water cooling rackmount gear has always been a challenge, not so much finding blocks or even routing the tubing but rad fitment and airflow challenges. By stacking radiators in multiples with a front-to-back airflow it can be done, especially if noise is not a concern and cooling performance is priority.

 

Maybe the next test we can see a Quad Xeon build water cooled or a GPU server with 8 Quadros under water. A set of 6 or 7 80mm thick 360 rads would definitely have no issues handling the heat of that many blocks. You also have the option to split the loop using dual pumps.

Link to comment
Share on other sites

Link to post
Share on other sites

TLDR;

Counter flow is best.

Dont double stack with a pressure drop zone between rads.

Warmed air is less efficient at cooling.

 

...duh,

 

😀

 

Woulda been nice to see a comparison of how all those rads would have performed un-stacked vs stacked. Just to really emphasis the point that, if at all possible, dont stack or recirculate warmed air.

CPU: Intel i7 3930k w/OC & EK Supremacy EVO Block | Motherboard: Asus P9x79 Pro  | RAM: G.Skill 4x4 1866 CL9 | PSU: Seasonic Platinum 1000w Corsair RM 750w Gold (2021)|

VDU: Panasonic 42" Plasma | GPU: Gigabyte 1080ti Gaming OC & Barrow Block (RIP)...GTX 980ti | Sound: Asus Xonar D2X - Z5500 -FiiO X3K DAP/DAC - ATH-M50S | Case: Phantek Enthoo Primo White |

Storage: Samsung 850 Pro 1TB SSD + WD Blue 1TB SSD | Cooling: XSPC D5 Photon 270 Res & Pump | 2x XSPC AX240 White Rads | NexXxos Monsta 80x240 Rad P/P | NF-A12x25 fans |

Link to comment
Share on other sites

Link to post
Share on other sites

15 minutes ago, xnamkcor said:

Now try parallel water flow?

with parallel flow, the flowrates would divide up based on number of radiators, decreasing their heat output individually, if one radiator can do the trick, then multiple in parallel would definitely work, it wont do much different, actually the heat transfer might be worse due to lower turbulence on the water side but with the higher surface area you probably wont notice it 

PC: Alienware 15 R3  Cpu: 7700hq  GPu : 1070 OC   Display: 1080p IPS Gsync panel 60hz  Storage: 970 evo 250 gb / 970 evo plus 500gb

Audio: Sennheiser HD 6xx  DAC: Schiit Modi 3E Amp: Schiit Magni Heresy

Link to comment
Share on other sites

Link to post
Share on other sites

nice work!

 

but in my opinion they have to measure the flowrates and the air temperature too.

so they check the heat input and output.

Link to comment
Share on other sites

Link to post
Share on other sites

@ColinLTT
Last time I honestly put forth examples indicating that stacking radiators like this isn't optimal.
And using some simple baffles, and giving each rad its own airflow is a lot more efficient.


And there is ample room for it even in a rather slim 2U server...

Spoiler

 

So here is an image of 3 different implementations for a multi radiator 2U server:
image.thumb.png.40bdd929d8f8439df15d4d8698a42815.png

 

Now, no radiator will heat the air to the liquid temperature of the loop.

But with each radiator the delta temperature will be lowered by X %. (The exact value of X is dependent on the radiator you have, and this can be calculated if you have knowledge of air speed, loop temperature, and the air temperature before and after the radiator, loop flow speed can largely be ignored if sufficiently high. (like 5+ L/minute for most computer radiators since the temperature variation over the radiator will then be fairly marginal compared to the delta between air and rad.))

So lets say that these radiators are able to exchange 20% of the delta. Ie, if the delta between the air and rad is 10 degrees, the air will be 2 degrees warmer after passing through the radiator.

 

Then lets say we have a loop temperature of 45C and an ambient of 25C.

Then both mine and Corsair's implementations would see a 4 degree C temperature rise for the air.
LTT's implementation would see a 4 degree rise for the first radiator, a 3.2 rise for the second and a 2.56 degree rise for the third.


Since every radiator has its own set of fans, then we can largely expect air flow through each rad to be roughly the same in all implementations.
This means that LTT manages to take a volume of X air per second, and heat it with 9.76 degrees.
Corsair manages to get to 4 degrees increase for twice the volume.
And mine is also 4 degrees but for three times the volume.

Now since we need the same amount of energy to heat X volume of air by Y degrees as we need to heat half of X volume by 2x Y degrees, then we can convert both mine and Corsair's results to be comparable to the LTT solution.

Ie, Corsair gets to 8 degrees, and I get to 12.
While LTT is at 9.76
For using the same amount of radiators as I do, LTT gets only 81.3% as much heat out of their system. (Ie, they either need more fan speed, or can't run as power hungry components.)

Now, Corsair gets beaten by LTT here to be fair, but if we rerun the numbers but say that our radiators are 50% efficient.
Then LTT gets 10 + 5 + 2.5 = 17.5
While Corsair gets 10 + 10 = 20
And I get 10 + 10 + 10 = 30

If the radiators are 75% efficient.
Then LTT gets 15 + 3.75 + 0.9375 = 19.6875
Corsair gets 15 + 15 = 30
And I get 45....

 

Still not tested though.

Not that its hard, just use some cardboard, some duct tape... And prove it for yourself.
Also, would be nice to get the test data from the video. Or at least for a given row of data points and ambient temperature during that test. From there, one could calculate a rough "efficiency" figure for the radiators at the given airflow.

I would also have to say that you can buy I2C temperature measuring chips that have a resolution in the 0.05 degrees C or better, so NTC termistors is kinda old school to be fair. (absolutely nothing wrong with them though.)

Edit:
Honestly, we could calculate the radiator efficiency by only knowing the inlet and outlet temperature of the two first radiators. We don't even need to know ambient air temperature. As long as we can make the assumption that all radiators have the same efficiency, and air/water flow rate through them as well.

Link to comment
Share on other sites

Link to post
Share on other sites

https://create.arduino.cc/projecthub/user1590239/wind-tunnel-water-sensors-28cf12
 

You are not authorized to access this page.
 
Link to comment
Share on other sites

Link to post
Share on other sites

I really enjoyed this, great work. A really good mix of data and ltt made it (and it's predecessor) fun to watch.

 

If you ever decide to do a follow up it would be really interesting to see the heat rejection from each radiator. (Q=MDOT.cp.∆t). Although you'd need a fairly accurate mass flow rate sensor for the coolant.  If you really wanted to geek out you could compare the estimated power draw as reported by the system for the CPU and GPU and compare it to the block heat rejection and then the pump and rads. Why if you had enough data channels you could even measure the air mass flow rate using a Maf from a car or similar and work out the radiator efficiency....

 

Aaand I've fallen down a rabbit hole. Pretty sure there's an xkcd for this.

 

Anyway thanks!

Link to comment
Share on other sites

Link to post
Share on other sites

with the min and max temperatures of 1-radiator setup go above and below the stacked 2-radiator setup — it is easy to “proof” both opposite opinion sides. And we are talking here 40 °C vs 38 °C. What was the reason to say “I'm wrong” ?

graph from video at 9-40.jpg

Link to comment
Share on other sites

Link to post
Share on other sites

Could we get the temperature data of ltt minecraft server built that was eluded to at the end of the video? Im really intrested in how it performend

Specs: CPU: I5 6600K (4.5 GHZ), GPU: RX 480 (stock), Mobo: MSI Z170A tomahawk AC, RAM: Corsair 16GB drr4 2600, CASE: NZXT S340, storage 240GB crusial SSD and a 1TB WD HHD

Link to comment
Share on other sites

Link to post
Share on other sites

This video was great, really leans on all of the strengths and know-how of the channel.

I would love more videos with this tone. And Arduino video series?

 

Bleigh!  Ever hear of AC series? 

Link to comment
Share on other sites

Link to post
Share on other sites

5 hours ago, fulminemizzega said:

@ColinLTTWhy use hardware trimmers instead of software calibration? Nice job anyway.

Main advantage of trimming before the ADC is that we can get better resolution in our measurement.
Since we don't need to waste ADC resolution on trimming.

Now, if one also were to toss in an op-amp to do both amplification and a bit of level shifting, then they could get a lot more resolution out of the termistors.

Considering that the ADC used were 12 bit, and the temperature range they measured were around 25-40 C (15 C delta)

Then theoretically, the max resolution should be 3.66 mC. (0.00366 C per step.) Though, this is "optimal" and noise would limit us before this, nor did they go to this overkill degree.

Link to comment
Share on other sites

Link to post
Share on other sites

8 minutes ago, Nystemy said:

Main advantage of trimming before the ADC is that we can get better resolution in our measurement.
Since we don't need to waste ADC resolution on trimming.

Now, if one also were to toss in an op-amp to do both amplification and a bit of level shifting, then they could get a lot more resolution out of the termistors.

Considering that the ADC used were 12 bit, and the temperature range they measured were around 25-40 C (15 C delta)

Then theoretically, the max resolution should be 3.66 mC. (0.00366 C per step.) Though, this is "optimal" and noise would limit us before this, nor did they go to this overkill degree.

Did you just use millicelsius? I like it.

Link to comment
Share on other sites

Link to post
Share on other sites

4 minutes ago, xnamkcor said:

Did you just use millicelsius? I like it.

Yes, its a rarely used way to describe temperature.
And honestly, its really hard to measure such small temperature differences in practice.
Since thermal resistance, mass and gradients becomes a major impact to all measurements.
And even the sense current used for making the measurement can greatly skew the answers by a few mC.

Just think of the troubles measuring µC or even nC!
Or all the people thinking one is talking about Coulombs. (1 volt Coulomb is 1 Farad. And has nothing to do with temperature btw.)

Link to comment
Share on other sites

Link to post
Share on other sites

2 minutes ago, Nystemy said:

No I didn't use µC, I used milli-Celsius. But you were close.

I've already corrected it, sorry.

Link to comment
Share on other sites

Link to post
Share on other sites

Just now, xnamkcor said:

I've already corrected it, sorry.

No worries.
I corrected mine as well.

Link to comment
Share on other sites

Link to post
Share on other sites

i like your units, but i think the more correct one for temperature differences is K [Kelvin] or at least you write °C (don't forget the little circle) ^_^

 

With the trimmer he only calibrate the offset of the thermistors in one point of the temperature curve. i don't see why you can't do this with the software too.

Link to comment
Share on other sites

Link to post
Share on other sites

52 minutes ago, Monk.ey said:

i like your units, but i think the more correct one for temperature differences is K [Kelvin] or at least you write °C (don't forget the little circle) ^_^

 

With the trimmer he only calibrate the offset of the thermistors in one point of the temperature curve. i don't see why you can't do this with the software too.

You can do it in software, no problem there - but doing it in hardware helps to reduce the data processing down the line. Garbage in = garbage out and so on. Also, a resistor is needed for the comparison in the equation anyway, so might as well tune the sensors in. 

Link to comment
Share on other sites

Link to post
Share on other sites

I have two concerns about the code:

  • You can run a running average in the arduino code,  instead analog reading 4 times per second, you maybe read more times per second avoiding the noisy read. https://www.arduino.cc/en/tutorial/smoothing
     
  • Even though the Steinhard equation have a well known constants, its recommended run the equation in the post processing (with excel, python, etc), instead in the arduino code. Why? If you have a bug in the code will lose precision trying recover the "correct data" or in the worst case escenario you will need rerun the test.

After that, nice analysis and really good plots.

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×