Sticking a TEC between GPU die and waterblock?

[yoadknux]

Hi guys,

I currently watercool my 1080ti using a NZXT Kraken X41. It has good performance but I'd like to take it one step further. I knew about the Thermoelectric effect for a while but didn't really think about utilizing it. I recently saw a video and the idea is that the TEC is a voltage-powered device composed of a small block that has two surfaces, a hot one and a cold one, and the idea is that the temperature difference between the two surfaces is kept constant. So what I want to do is to put the cold side on the GPU die, and the hot one on the waterblock.

Has anyone tried this sort of thing here? Do you have experience with it? Does it work or fail horribly?

Thanks

 

[bignazpwns]

Horrible idea. They are extremely inefficient and in general just suck. Power consumption alone is not worth it. That water is pretty maxing out your card. You won't see any improvement worth the investment using it

 

[Zerileous]

To add to the list of cons: Unless you have a controller to cycle it off and on you'll have to deal with sub ambient temps, frost, frost melting under load, water everywhere, condensation everywhere too. You'll also need way more rad to cool it because as bignazpwns mentioned, it's really inefficient so it will be adding way more heat to your loop than it will take away from the card. And if you exceed it's thermal capacity it turns into a giant insulator.

You want better than ambient liquid cooling, look at these http://www.performance-pcs.com/water-chillers

 

[yoadknux]

Haha, well didn't know they sucked that much. Thanks for the comments.

 

[Super Nade]

Haha, well didn't know they sucked that much. Thanks for the comments.

The point being made is TEC efficiencies are only ~ 1-5% at high thermal load, i.e. large temperature differentials. This is because in the TEC material, a large fraction of electrical power drawn is dissipated as useless joule heat. The point is, you need a good high current PSU, with low ripple (preferably dedicated) and possibly a chiller to make this work. Setting aside poor energy efficiency, for optimum usage, you are looking at significant changes to your existing power/cooling system. We use TECs all the time in our lab to control laser diode temperatures (< 50 W), but that is low power compared to a GPU.

Here are a few good resources to understand TEC technology.

A practical showcase of inefficiency:
https://rimstar.org/science_electronics_projects/peltier_effect_module_cooling_efficiency_test.htm


Basics:
https://www.marlow.com/resources/thermoelectric-technology-guide
https://www.meerstetter.ch/compendium/peltier-elements#COP
https://www.europeanthermodynamics....ermoelectric cooler module in a system v2.pdf

Advanced reading:
https://phys.org/news/2019-01-thermoelectric-material.html
https://arxiv.org/ftp/arxiv/papers/1607/1607.03357.pdf

Simple calculations:
https://www.ijstr.org/final-print/may2016/Improvement-In-The-Cop-Of-Thermoelectric-Cooler.pdf

MATLAB Simulation:
https://www.mathworks.com/help/phys...-peltier-device-as-thermoelectric-cooler.html

 

[yoadknux]

The point being made is TEC efficiencies are only ~ 1-5% at high thermal load, i.e. large temperature differentials. This is because in the TEC material, a large fraction of electrical power drawn is dissipated as useless joule heat. The point is, you need a good high current PSU, with low ripple (preferably dedicated) and possibly a chiller to make this work. Setting aside poor energy efficiency, for optimum usage, you are looking at significant changes to your existing power/cooling system. We use TECs all the time in our lab to control laser diode temperatures (< 50 W), but that is low power compared to a GPU.

Here are a few good resources to understand TEC technology.

A practical showcase of inefficiency:
https://rimstar.org/science_electronics_projects/peltier_effect_module_cooling_efficiency_test.htm


Basics:
https://www.marlow.com/resources/thermoelectric-technology-guide
https://www.meerstetter.ch/compendium/peltier-elements#COP
https://www.europeanthermodynamics....ermoelectric cooler module in a system v2.pdf

Advanced reading:
https://phys.org/news/2019-01-thermoelectric-material.html
https://arxiv.org/ftp/arxiv/papers/1607/1607.03357.pdf

Simple calculations:
https://www.ijstr.org/final-print/may2016/Improvement-In-The-Cop-Of-Thermoelectric-Cooler.pdf

MATLAB Simulation:
https://www.mathworks.com/help/phys...-peltier-device-as-thermoelectric-cooler.html

Thanks for the detailed comment. It's an interesting subject. Why do you need both a chiller and a TEC? Is it possible to think about the TEC in your case as a really good thermal conductor? What exactly do you cool in the laser, is it a diode, a gain medium, something else? I know that for example in nonlinear optics it's very important to fine-tune the temperature, and I wonder if this is done with TEC or something else.

 

[Super Nade]

Thanks for the detailed comment. It's an interesting subject. Why do you need both a chiller and a TEC? Is it possible to think about the TEC in your case as a really good thermal conductor? What exactly do you cool in the laser, is it a diode, a gain medium, something else? I know that for example in nonlinear optics it's very important to fine-tune the temperature, and I wonder if this is done with TEC or something else.

Laser wavelengths can be tuned by varying temperature (slow), current (fast), piezo (cavity length variation) and optical injection feedback (send light back into the laser) tuned. In general, you try to keep the temperature of a laser constant. We use lasers ranging from 30 W to 300W CW. A laser is only ~50% wall-plug efficient, so the remaining energy gets dissipated as heat. If the temperature is allowed to run away (called thermal runaway), irreversible electro-migration damage is possible. This "ages" the diode rapidly, leading to catastrophic failure.
https://www.thorlabs.com/NewGroupPage9_PF.cfm?ObjectGroup_ID=1832

In the diode lasers we use (fiber lasers), tuning is possible over a small wavelength range by temperature changes. The key here is to change the cavity length by heating/cooling, which are diffraction gratings etched on a fiber (fiber bragg gratings or volume bragg gratings). We are building laser amplifiers which require stable wavelength to keep power fluctuations small, so temperature control is necessary.

If you are interested in fiber lasers check out:
https://www.rp-photonics.com/fiber_lasers.html

I don't post here a lot, like I used to. But when I do, I try to make it count.