PDA

View Full Version : new improved pelt technology


g.l.amour
04-02-2003, 02:25 AM
http://www.newswise.com/articles/2003/3/FRIDGE.ACS.html

quote:
"If we are fully successful, we can imagine the possibility of replacing most of the mechanical refrigerators and air-conditioning systems with CFC-free, solid-state, no-moving parts, and therefore reliable, electronic heat pump technology," Venkatasubramanian says.

they think that with further development the element can be made 2 to 3 times more efficient than conventional pelts.

gmat
04-02-2003, 07:50 AM
"active-cooling power densities approaching 700 W/cm2"

wow ! is that an April's fool ??!!

If not, i just hope the price won't be *that* high...

g.l.amour
04-02-2003, 08:03 AM
shit, now that u mention it, lets hope indeed that its not an april's joke.

winewood
04-02-2003, 09:58 AM
Even with this cooling capacity... would it eliminate the need for water cooling to transfer the heat away? I wouldn't think it would. Basically it would act like a super efficient pelt?

g.l.amour
04-02-2003, 12:17 PM
sure u would. it is still a heat pump; let it be impossibly perfect, it would still have to take away the cpu heat.

airspirit
04-02-2003, 01:07 PM
If they are that efficient, though, couldn't you possibly get away with high-end air cooling of the pelt? I mean, it's possible with the 50-60W variety today, but if this was 2-3X more efficient, wouldn't a 120W pelt be in the realm of possibility? And if so, wouldn't that be an absolute dream for the overclocking community? The only difficult step would be insulation, and you wouldn't have to get over the less experienced overclocker's fear of watercooling in order for them to take a step up to the next level of cooling.

airspirit
04-02-2003, 01:08 PM
The article was dated 25 March, btw. I don't think it's a 4/1 joke.

airspirit
04-02-2003, 01:12 PM
Oh, and reading the article, it looks like they currently have pelt efficiencies, just packed into about 25% of the size. Imagine being able to put four of these bastids on one waterblock ... can anyone say "Prometia Killer?"

Oh, and can anyone say "Insane Power Bill?"

MeltMan
04-02-2003, 03:14 PM
I can!

"Insane power bill!"

gmat
04-03-2003, 04:47 AM
700W per cm^2... since it works like a pelt (heat pump) you *will* have to use watercooling (with a goooood rad !) on the hot side, or else you'll melt everything.
I see a few problems:
- even if the device is 1cm^2 (very roughly the size of a CPU core) -> it will need 700W of power. The prices of 700+W PSUs are, hhmmmmm, painful
- the 'cold' side may drop below the 130°K which is the point where semiconductors (aka transistors in CPUs) stop being 'semi' conductors, and start conducting fully. Which is, for a CPU, not good.

My guess is that we'll leave that technology together with helium phase change cooling... not adapted to our budgets / needs.

Since87
04-03-2003, 08:06 AM
Originally posted by gmat

I see a few problems:
- even if the device is 1cm^2 (very roughly the size of a CPU core) -> it will need 700W of power. The prices of 700+W PSUs are, hhmmmmm, painful
- the 'cold' side may drop below the 130°K which is the point where semiconductors (aka transistors in CPUs) stop being 'semi' conductors, and start conducting fully. Which is, for a CPU, not good.


There is no need to operate existing TEC's at their Vmax spec. In fact very few people (and a foolish few at that) do so.

Assuming these new devices behave similar to exsisting TEC's, it is only a matter of operating the device at an appropriate voltage for the application.

airspirit
04-03-2003, 10:27 AM
Preeeecisely. I'm not advocating shooting for 10 deg K, I'm advocating shooting for Prometia range temps, which should be achievable with these things. Using two or three of them at "low" power would make it feasable to set up a redundant 350W PSU array to cool your processor, running each pelt at, say, 100-150W. If one self-destructs, or a PSU goes out, you still would have others to fill it's cooling duty until you could replace it. Since 350W PSUs can be had for $20 a piece, this could be a VERY cost effective way to get subzero temps.

Of course, this type of plan would require making a resistor enhanced baybus to regulate the voltage/amperage going through each pelt unit, and that would require some nice cooling in and of itself. I would suggest that two standard mid-tower cases side by side (or a YY cube) could contain all the necessary components to do this.

gmat
04-03-2003, 01:38 PM
- The 350W PSUs you cite do not provide in any way 350W of power on a single rail. If you want 12V/350W (for example) you'll pay a lot more.
- In my previous post i carefully chosed the 'power' term and not 'voltage'. If you want, say, 100W of cooling, with 50% of inefficiencies (not far away from reality) you must provide 150W of power. Even when using a 200W rated device.
- if you happen to use one of those devices at 10% of its rated power, how does it behave ? Is it any better than an 'usual' pelt ?

The strong point of this new device is thermal density. It's clearly beyond the best peltiers i've seen or heard of.
Another strong point seems to be efficiency.
But i'll take the example of helium phase change cooling: it's possible, works wonders, but aint it a tad expensive ?
Given the specs of that technology, i fear the price would make it useable by army, aerospace, labs but not PC enthusiasts...

airspirit
04-03-2003, 02:31 PM
Notice that I said that I'd set them up to run at 100-150W via resistor arrays, each on a seperate cheap PSU? I fully understand what you're saying ... read my post again.

When you have three or four of them cooling at around 80-100W (actual), you can get your CPU very cold (easily subzero), you still will take up less space on the block than a standard pelt, and your system will be very failure resistant (even if 3 of 4 blow out, for example, one of them will keep your system from going nuclear until you can shut it off, even if it happens during the middle of the night and you don't catch it until the next morning).

The size per cooling ability is what will make this work, even if each individual mini-pelt does no more than a standard pelt does ... the advantage is the amount of them you can fit feasably per block.

g.l.amour
04-03-2003, 04:14 PM
of course it will be ridiculously expensive atm, but isn't any new technology? if there comes a need for large scale production then we will see some nice prices. i mean, look at that sony blue ray dvd thingy costing approx 3600$, lets see what that will cost in 2 yrs.

it will prolly turn out nothing, but then again, maybe then those stirling coolers will be the "hot" things.

Since87
04-03-2003, 11:16 PM
Originally posted by gmat

- In my previous post i carefully chosed the 'power' term and not 'voltage'. If you want, say, 100W of cooling, with 50% of inefficiencies (not far away from reality) you must provide 150W of power. Even when using a 200W rated device.
- if you happen to use one of those devices at 10% of its rated power, how does it behave ? Is it any better than an 'usual' pelt ?



My impression of these devices, based on the link above, is that they are similar to existing TEC's in behavior but much denser.

Based on that, a CPU sized module may behave similarly to four 172 Watt TEC's.

Below are some simulations of a system of four 172 Watt TEC's in parallel, at voltages available from a PSU.

I've chosen 0.22 C/W for the hotside parameters based on a White Water and reasonable radiator, etc.

I've chosen 0.1 C/W as what I believe to be a fairly conservative thermal resistance between the CPU and the coldside of the device.

CPU power dissipation is set at 100 Watts.

First off 12V:

http://uffish-thought.net/wc-gifs/4drift-12.gif

In this simulation CPU temperature is 3.9C with a TEC power consumption of 247 Watts. This is vastly better than anyone is is doing with a TEC now with that much power consumption.

Next consider the setup powered at 5V:

http://uffish-thought.net/wc-gifs/4drift-5.gif

Only 45 Watts power consumption and the CPU is still 4C below ambient.

Finally 3.3V:

http://uffish-thought.net/wc-gifs/4drift-33.gif

A mere 21 Watts still maintains the CPU near ambient.

They may not be affordable anytime soon, but this technology may be absolutely necessary for Moore's law to continue to hold true.

Les
04-04-2003, 07:57 AM
Since87.
Not sure but suspect your Rh may be a little optimistic.
To my understanding Kryotherm calculates using Rh per module.
Maybe Rh=0.88c/w is more realistic.
No big deal but worthy a moment's ponder.
This no way affects your low power consumption conclusion.

Going to the pub.Back Sunday

Since87
04-04-2003, 08:14 AM
Les,

I was treating this as a simulation of a hypothetical version of one of these new devices. According to the document, one of these devices can pump 700 Watts through 1 cm^2. Thus the use of four 172 Watt TEC's (688 Watts) to simulate the electronic aspects of a die sized version of one of these.

The document said the prototype was, "about the size of a large postage stamp". I'm presuming that one could be manufactured with the dimensions of a CPU die.

Edit: Nope, I was wrong, you are right. The numbers don't make sense unless the kryotherm software divides the C/W by four for the four TEC's.

Redoing the simulation with operating voltage of 5V and hotside C/W of 0.88, and Coldside C/W of 0.4 yields:

Tob = 39.5 C
W = 42
Qh = 124.5

(I'm at work without graphic tools so I can't show the pictures.)

Optimal cooling is at around 7V but 7V only gains 1C over 5V.

Fairly pointless to use a design like this. The waterblock is too much of a bottleneck. Power consumption is not the big issue though.

Unless the predicted 2-3 gain in efficiency can be achieved these don't appear so useful.

Maybe combined with cheap diamond heatspreaders and methanol phase change... :D