Will low power CPUs bring back TECs?

[Uncle Jimbo]

AMD Deneb and Intel i7 CPUs are showing much lower power at load than previous generation, even when set to high OC. Does this open the door to direct TEC cooling again?

[Uncle Jimbo]

A friend has an AMD Phenom II which he overclocked to 3.8G using only 1.45V (stock is 1.35V) core. On stock cooling, he saw full load temps under 50C. The total dissipation was under 75W at that OC (as calculated from heat sink rise). These chips have clocked to over 6G using Liquid Nitrogen... it would be intereasting to see what a direct TEC cooling setup could do. A water cooled 12730 could easily take the chip sub-zero with reasonable efficiency at that power level, possibly allowing better than 5G speeds. Intel's i7 is not quite as efficient, but still better than the previous generation, showing around 150W TDP at 3.8G OC - so a good setup might get the chip to 10C and open up more headroom.

Even a stock product like the Swiftech MCW6500-T would take the Phenom II subzero with around 125W in, which gets to a CoP near .6, not bad for a TEC solution. Total heat load is only 200W.

This might be a new day for TEC cooling of CPUs.

[NaeKuh]

no jimbo,

we undershot the head load on the i7.

these arent cool running chips... infact id rank them near B3 kentsfield, and possibly greater depending on overclock.

130W TDP @ stock is no joke.

[Vinas]

Yeah problem is the TEC likes to heat up under load and cause the TEC to swing up in temps so drastically. So probably not in the neighborhood of LN2... But I agree with you Jimbo on the fact that higher efficiency will bring back TEC eventually, even if just a "better than water" solution.

The quads might take a year or two but eventually, I think they'll be good candidates for TEC. By then though we'll have even more cores to cool down. =P

[Uncle Jimbo]

Well I'm impressed enough by Phenom II heat load (maybe 100W at 4G) to give a shot at TEC cooling it. Using a 12730 at 5V, I can get a 20C differential moving 100W, and better yet, at that low voltage, the cold side will not drop much as heat load drops, because the higher dT needs more voltage. That means I effectively control dT to about 20C just from the physics of the setup.

Draw is less than 8A so I can run it off of a molex from the PSU. Total heat load (assuming 100W CPU load) is 140W. If I use a TRUE cooler with 110 CFM, heat rise will be 10C, so I get the chip to 10C below ambient, which gives nice OC room without big concerns about condensation.

Some posers are hitting 4.4G and higher with that chip with cooler temps at 15C, so this might be a relatively cheap and easy way to get some big overclocks without much fuss.

[Ozzfest05]

TEC's are really really really really ineffiecient and dont perform well at all on current cpu's all the old school TEC coolers have long since changed their ways because they just dont cut it anymore and its becomes too much work and power draw for so little gain.

Use air then water then phase skip the TEC

[flak-spammer]

I think if there were TEC coolers that focused on efficiency in numbers rather than total cooling power. I mean think of 45 watt TDP processors we could probably cool them on TECs at low input voltages such that the efficiency would be quite high and input power would be lower. TECs may be very inefficient but at the same time they are much smaller than comparable sub-ambient cooling technology, think phase-change cooling. So if there were some smart designers I'm sure that TEC's could be used to cool low power CPUs. Good example would be the mobile style quad cores.

[SNiiPE_DoGG]

TEC's are really really really really ineffiecient and dont perform well at all on current cpu's all the old school TEC coolers have long since changed their ways because they just dont cut it anymore and its becomes too much work and power draw for so little gain.

Use air then water then phase skip the TEC

you realise you are telling the local forum TEC expert not to use TECs LOL!

he knows more than all of us combined IIRC

[Ozzfest05]

you realise you are telling the local forum TEC expert not to use TECs LOL!

he knows more than all of us combined IIRC

Oh im not saying TECs are out of the question I mean if you have all the equipement then great but a core I7 I mean come on it wouldnt be worth it its just my opinion I have friends the TEC cooled for the longest time and now they tell me its not worth it at all again im not saying im experienced in TEC cooling im saying that with todays processors its not worth it


EDIT: I didnt mean to come off as ignorant or anything I do have all the respect for someone who can pull off a good TEC setup.

[Uncle Jimbo]

TEC's are really really really really ineffiecient and dont perform well at all on current cpu's all the old school TEC coolers have long since changed their ways because they just dont cut it anymore and its becomes too much work and power draw for so little gain.

Use air then water then phase skip the TEC

There are three things wrong with your statement -
1 - TEC's are really really really really ineffiecient
I will be using an operating range that will give CoP between 1.3 and 2 - that's about as good as a phase setup. With power draw under 80W, there is also no phase setup that can compete for power use. I don't think phase is really really really really inefficient - TECs aren't either, unless you are above 35% Imax.

2 - I can buy all of the parts to do this (including the cooler) for under $80. Check out what the minimum entry point is for phase. If you already have a decent cooler you can add TEC for $20 or so.

3 - The design and setup is really simple - temp range is high enough to avoid condensation (with a little care) but low enough to add a lot of OC range. Using an off the shelf thermally controlled fan, I'll be able to control cold side temps just by letting the hot side run up during low power CPU use. So even beginners can do it without much risk or effort.

If you look at the title of the thread, you will note the words low power CPU - that pretty much eliminates i7. However, e8600 would do really well under this setup, as well as Phenom II

[flak-spammer]

1 - TEC's are really really really really ineffiecient

See I think the real problem is that people think TEC's are really really inefficient. I suppose if people are looking for the MOST and absolute MOST cooling power then they are. However with a lower voltage I think they become within a range to compete with Phase change. I suspect that people get the idea that phase change is really more efficient than it actually is. Compressors actually are quite a bit less efficient than people get the idea of. Obviously TEC's are quite inefficient at 100% U-max, I'm pretty sure manufacturers do not recommend running then at 100% more like 70-80%.

[ElMoIsEviL]

TEC's are really really really really ineffiecient and dont perform well at all on current cpu's all the old school TEC coolers have long since changed their ways because they just dont cut it anymore and its becomes too much work and power draw for so little gain.

Use air then water then phase skip the TEC

I used to hear this line all the time as I sat there with a Q6600 G0 at 4GHz idling at -17C and loading around 35C using a simple MCW5002-775T.

It would make me cringe anytime I heard this.

TECs can still be applied. And some folks have come up with products using several smaller TECs to achieve better than water results.

Yes, they consume a lot of power but I've found them to be quite useful.

http://forum.xcpus.com/gallery/d/3011-1/test_1.jpg
http://forum.xcpus.com/gallery/d/3284-1/hot_day.jpg

Of course I was also using a slight TEC chiller at the time. It cost nearly nothing to build.

[Ozzfest05]

So wait Jimbo say im running an 8600 which I am cooled with a apogee gtz your saying and 80w pelt would be sufficient to cool it near 0 or just below ambient under idle, I have a pelt I could use on this system I just didnt know if I could hook the pelt directly into my hx1000w without causing problems I love my psu too much to hurt it

[jcool]

AMD Deneb and Intel i7 CPUs are showing much lower power at load than previous generation, even when set to high OC. Does this open the door to direct TEC cooling again?

Hey Jimbo,

that may hold true for AMD but the i7 sure are power-hungry beasts, especially above 4Ghz and with HT enabled. I can easily bring my Phase unit down if I set 1,5V+ Vcore and run linpack for a bit (my unit has a thermal capacity of around 280W) so go figure

[Uncle Jimbo]

Hey Jimbo,

that may hold true for AMD but the i7 sure are power-hungry beasts, especially above 4Ghz and with HT enabled. I can easily bring my Phase unit down if I set 1,5V+ Vcore and run linpack for a bit (my unit has a thermal capacity of around 280W) so go figure

Phase is the way for i7. After good results on e8600 I did some power testing on i7 965 - there is no TEC that can cool that sucker with OC at load. I would not recommend TEC for any CPU that needs more than 150W at max OC, 180W with the TEC at absolute crazy top. And for anything over 100W, you have to WC the TEC hot side.

But with the much lower thermal profiles of the C2D, many of them look like TEC would work. Those chips also give some nice OC even when the cooling is only a little below ambient - I think they have plenty of life for those who don't have the $$$ to do i7 at full bore.

[jcool]

As a matter of fact, I will soon change the phase (pun intended) for a custom waterchiller unit - less noise and electricity costs, more flexibility (using a variable RPM compressor, heat loads between 300 - 800W can be accomodated easily at 150-350W power drain). But this is another story and shall be told another time

[Uncle Jimbo]

So wait Jimbo say im running an 8600 which I am cooled with a apogee gtz your saying and 80w pelt would be sufficient to cool it near 0 or just below ambient under idle, I have a pelt I could use on this system I just didnt know if I could hook the pelt directly into my hx1000w without causing problems I love my psu too much to hurt it

The GTZ is a great block and gives you a lot of flexibility in mounting. With a thermal resistance of only .04 to water at 2 GPM, it can handle large power with little heat rise. Let's look at how it would work with the 50mm '340W' sold on ArcticSpider
340 Watt Peltier

and the 62mm 12630 sold on eBay as
Gigantic 62mm 545 Watt Thermoelectric Peltier Cooler

I set up an e8600 at 4G 1.26 vCore and measured under 80W heat load. My guess is you want to run higher clock, and if you give me the current numbers you are getting, I can do a better calculation. But lets target 100W as your heat load just for discussion.

Running the 12730 at 5V will get you well under ambient with low power draw and plenty of headroom. With that block, you might end up 8-10C under ambient at the IHS, (maybe 15-18C actual) even at load, with a quick and easy installation.

The 340W TEC is a 50mm unit, which makes for easier mounting - it can be directly substituted in Swiftech's MCW6500-T TEC block, for example. With voltage at 12V, you will draw about 6A - so you could still power it from a MOLEX peripheral connector, but using the 12V rail. Total heat load to the block will be 76W from that TEC heating plus the 100W or so CPU cooling, which will give a rise to water of 7C with your block.

Actual rise to ambient depends on your rad setup but good rigs are showing .02 or better. So total rise to ambient might be 10C or so. The TEC will provide around 25C drop at those power levels - so you should see 15C below ambient at the IHS (10C in 25C ambient). At those levels, you will get CoP of about 1.3 good efficiency for a TEC.

Since this operating point is only 30% of Imax, you can maintain decent efficiency at increased power levels. For example, at 16V, you get an additional 5C drop, but rise only goes up 2C, so you pick up another 3C - maybe 8C actual in 25C ambient - at the IHS. Efficiency at that level is CoP of .9 and power needed is 8A - still safe with the connectors on your PSU, but since you can't get 16V, you would need a special supply.

You can use the 12730 eBay TEC at 5V if you don't need a big drop, but it does not do as well at higher power levels. A Meanwell SP-320-7.5 can be set to any voltage between 6 and 9 volts. With a 12630 TEC, and voltage at 8V, you will draw about 15A. Total heat load to the block will be 120W from that TEC heating plus the 100W or so CPU cooling, which will give a rise to water of 9C with your block.

Total rise to ambient might be 13C or so. The TEC will provide around 30C drop at those power levels - so you should see 17C below ambient at the IHS. At those levels, you will get CoP of about 1, not great efficiency but not bad.

Since this operating point is already at 50% of Imax, efficiency goes down fast at increased power levels.

[Ozzfest05]

extremely informative Jimbo but the 545 watt pelt would require a seperate power supply a rather large one no? the 320 meanwell wouldnt cut it ? thats where pelt cooling gets me is the seperat psu if it was a benching rig I wouldnt care by this is my 24/7 rig

[SNiiPE_DoGG]

there is a 600w meanwell....

[Uncle Jimbo]

extremely informative Jimbo but the 545 watt pelt would require a seperate power supply a rather large one no? the 320 meanwell wouldnt cut it ? thats where pelt cooling gets me is the separate psu if it was a benching rig I wouldnt care by this is my 24/7 rig

No - what I am saying is you can run the 545 pelt directly from a 5V MOLEX line and it won't even use much current. If all you need is to go a little below ambient, this should work fine. 8A won't be much of an additional load for any decent PSU. Your PSU won't even notice the difference.

I know lots of people run that TEC from a Meanwell at 14V or more but that's just a good way to heat up your house. If you want to run that TEC over 5V, it should never be used over 9V - the slight gains in cooling don't justify the other problems. At 9V it only draws 15A - that's only a total load of 135W, which is not big power.

Running that 545 TEC at 12V, 14V or any of the other high voltages is just from people who don't understand how to best use a TEC for efficient cooling. That's one of the reasons TECs have been labeled as inefficient.

[Uncle Jimbo]

there is a 600w meanwell....

The '545 watt TEC' draws 22A at 12V, 264 watts - well within the range of a Meanwell 320. There are not many uses for using that TEC at 12V anyway... maybe for brief peak load control if using PWM control.

[Ozzfest05]

so jim if you use a 12 or 14v pelt on a molew (the 545 pelt) it wont draw more current then the psu through molew can handle im using hx1000w and is there a better tec well I guess I mean smaller then the 545 watt something closer to the size of a cpu more like 50mm by 50mm

[flak-spammer]

I don't totally get what you are saying but I wouldn't go past 9 amps 10 tops on a molex connector that being said a higher wattage TEC operating at lower input voltages are more efficient than a lower wattage operating at a higher voltage. At the same time the lower wattage TEC will get colder but will be unable to move as much of a load. So anything above 9 amps I would not power from a computer PSU. I had my 136.8 watt ones going at 12v and the tails got pretty darn warm and that was only at 7.2a. I believe the max current for a molex is 9 or 10 amps. Hope that answered the question if not I'm sure Jimbo will say it better.

[Uncle Jimbo]

so jim if you use a 12 or 14v pelt on a molew (the 545 pelt) it wont draw more current then the psu through molew can handle im using hx1000w and is there a better tec well I guess I mean smaller then the 545 watt something closer to the size of a cpu more like 50mm by 50mm

Just be sure you run a 16V TEC like the 545 from the 5V line (RED). It will draw only 8 or 9 amps. If you hook it to the 12V line it will melt the Molex and probably blow your supply, so don't do that.

The 340W TEC from ArcticSpider is a 50mm by 50mm 36V peltier. It is probably a better part for what you are doing - right size and more efficient. Because it is a higher voltage part, you can run it at 12V from your PSU. the hx1000w should have no trouble running either of those TECS at these undervolted levels.

Be sure you read up on TECs a little. You can't just put the TEC between the CPU and block - it needs more tension. For this kind of power, you can use a fairly small 'sandwich' of a TEC and 2 copper plates, and that sandwich can just sit between the CPU and block, because the copper plates put the pressure on the TEC. Or look at the Swiftech design -
mcw6500-t1
They don't use a big enough TEC for e8600, but it is a 50mm design and well done aside from the power of the TEC. They actually mount the TEC directly to the block, using the bottom of the water chamber as the hot plate of the TEC.

This should be pretty easy if you use a little care, but it does require that you pay attention to pressure on the TEC and overall mounting. Also if you go much below ambient you need to watch condensation.

If you have the money, and want an easy installation, it might be worthwhile to buy the MCW6500-T and the ArcticSpider 340 TEC, and swap out the one that comes with the MCW6500-T. It is not as good a block as yours, but it's not bad.

That just requires unscrewing the mounting bolts and putting the new TEC in. Be sure you get the hot side facing the water block!!!

[Uncle Jimbo]

I don't totally get what you are saying but I wouldn't go past 9 amps 10 tops on a molex connector that being said a higher wattage TEC operating at lower input voltages are more efficient than a lower wattage operating at a higher voltage. At the same time the lower wattage TEC will get colder but will be unable to move as much of a load. So anything above 9 amps I would not power from a computer PSU. I had my 136.8 watt ones going at 12v and the tails got pretty darn warm and that was only at 7.2a. I believe the max current for a molex is 9 or 10 amps. Hope that answered the question if not I'm sure Jimbo will say it better.

You said it pretty well... the standard white Molex connectors are rated for only 8A. The ones on newer high end PSUs have a different design and are rated for 10A. Probably good to stay an amp under the max rating, so 7A for a standard Molex, and 9A for the newer contact design.