Peltier cooling

[CanaryInTheMine]

So, I've been thinking...

In order to push the cards and achieve stability with higher OC rates, VRM temps is what one needs to worry about more than GPU temps. (if you don't care about fan noise).  I'm assuming that you took care of the airflow already and that you have been able to apply necessary voltages (higher) to achieve high OC stability.

My question is, has anyone looked into Peltier cooling?

Seems it may be possible to fashion something (at least on certain cards) on the back of the card (if that's where the VRMs are emitting at least some of the heat).  Adding a heatsink over the thermoelectric cooler surface would also be possible but not necessary in order to remove heat from the board/VRMs, transfer to Peltier's surface and let appropriate airflow take it out the rest of the way.

If you understand what I'm talking about, please comment.  Is this feasible? viable? worth pursuing?  I'm planning on experimenting with this.

Has anyone experimented with this yet?

Thoughtful and useful comments are welcomed!

[1713278862]

1713278862

[1713278862]

1713278862

[1713278862]

1713278862

[casascius]

I'm familiar with Peltier cooling, I think it would be terrible.  Those GPU's put out far more heat than your typical Peltier cooler is made to handle, and THEN you need to dump the heat from the other side of the cooler - both the heat it took out, and the additional heat the cooler generates in the process of pumping the heat (since it consumes energy as well).  If you want to get fancy, water cooling is the way to go.  If you want to save energy, have that heated water heat your pool or hot tub or house or something.

[mrNightowl]

Direct cooling I don't think would work that well with a peltier.  Not to mention condensation issue. 

I would think your best bet with a thermoelectric cooling device would be used in conjunction with a liquid cooled system.

[CanaryInTheMine]

Can't use liquid cooling in a data center here...  It's a disaster waiting to happen when something bursts.

You don't think it would be able to remove 10-15 degrees? Anything that can take VRMs from 100* down a few notches might be useful in extending the lifetime...

[casascius]

Can't use liquid cooling in a data center here...  It's a disaster waiting to happen when something bursts.

You don't think it would be able to remove 10-15 degrees? Anything that can take VRMs from 100* down a few notches might be useful in extending the lifetime...

Can a 1 square inch Peltier cooler pass over 100W of heat (plus overhead), constantly, without melting?  Then sure, it'll work.

[CanaryInTheMine]

Can't use liquid cooling in a data center here...  It's a disaster waiting to happen when something bursts.

You don't think it would be able to remove 10-15 degrees? Anything that can take VRMs from 100* down a few notches might be useful in extending the lifetime...

Can a 1 square inch Peltier cooler pass over 100W of heat (plus overhead), constantly, without melting?  Then sure, it'll work.

It would have to be 1.6-2 square inches...

[mrNightowl]

I think it would be risky putting any size.  If the module is not evenly heated it will create condensation.

I seen a guy at a lan-party once that did use 2 large peltiers to create a mini air conditioning unit for his case though.  Not sure how much cooler it got in the case but it had a 5 inch fan blowing across copper fins that had been cooled to frost just about.  They are cheap enough to buy on ebay.  Could be a fun little project if your bored.

[Littleshop]

Can't use liquid cooling in a data center here...  It's a disaster waiting to happen when something bursts.

You don't think it would be able to remove 10-15 degrees? Anything that can take VRMs from 100* down a few notches might be useful in extending the lifetime...

Use low conductivity liquid.

Peltier will not work in this application for direct cooling.

If you want to have fun, use them to generate power on the secondary side of a water cooling system.

[CanaryInTheMine]

Thanks for your inputs!

I'll put this on the back-burner for now... When I'm bored I'll experiment.

[Apopfis]

Peltier cooling is terribly inefficient you put in 100W of electricity you get 50W of heat removal from the source at 10 degrees celsius difference. So now you need to cool 100+50= 150W of heat. Also you cannot cool any modern graphic chip with 100W peltier (the chinese cheap peltiers you find on ebay are marketed by their electric consumption watts). This thing scales up as terribly inefficient. http://www.customthermoelectric.com/tecs/pdf/19911-5M31-28CZ_spec_sht.pdf just look at that graph you find bottom left. It can move 400W of heat at zero degree temperature difference at the same time it's also taking in 694W of electricity which it is rated at. With same consumption it can remove 330W @ 10° C difference.

Ok now you need to be able to cool all that power. 330W + 694W = 1024W of heat coming from that system. If you can keep your cooling system at room temperature level for long periods of time maybe then you need to worry about condensation. But let me tell you when starting to play with peltier, condensation will be least of your problems. You really need direct watercooling for peltiers at that level, air is out of the question. I use this example to explain the problems to explain some things you need to understand your mileage may vary.

Also you need separate psu for that peltier 24Volts that can easily deliver required 28Amps thats around 700W atleast and safe margin on top of that.

Instead get window airconditioner and put the condensator into a box of water and run that to cool your equipment with water loop. Normal portable airconditioner around here (northern europe) take in 900W of electricity and through phase change remove 2600W of heat from the source.