TECs and fault-tolerant design.

[wbedard]

Having read through quite a bit of watercooling material over the recent months, I have noticed that many people are using the WB+TEC setup on their cores and are getting very good temperatures in doing so. While I have read a few cases where the TECs have failed and things have gotten correspondingly ugly, I have not read where anyone has come up with a fault-tolerant design to preserve the CPU if the TEC fails. My understanding is if that happens, then the TEC acts as an insulator which causes the CPU to undergo heath death in short order. I know that software such as Motherboard Monitor can shut things down if it senses a dangerous temp but is that currently the "state of the art" solution?

[bigben2k]

search for OCP.

(Welcome!)

[wbedard]

Thanks bigben2k. Reflecting for a moment, I had read about that feature some time ago...

If I remember correctly, Intel has had it on their CPUs for a while and AMD recently specified it for their systems, as well. I think, initially at least, it was added as a module to the motherboard and I guess most of the 'board manufacturers have added it by now. Do any exceptions come to mind? Does AMD build it into their CPUs yet.

I guess my main concern then is how OCP operates in a 2x CPU system. Is it correct to say that if one CPU gets too hot that the entire motherboard is shut down?

Thanks again for reminding me about this...:>)

[Can O' Beans]

I can't remember where(here?), but I've seen someone use a "Thermal switch" hooked up to the Green ( i think) wire of the power supply.

You could put the switch on the cold plate so if the temp of the CPU/coldplate got too hot, the switch would imediately shut the whole system down.

[warpath]

the problem with turning it off via MBM is that the speed at which readings are measured is far too slow to effectively shut-down before heat death occurs.

[GigaFrog]

"Thermal switch" ??
That's me !!!

About a year ago, I proposed this solution:
The green wire carries the "PS_OK" signal, which the mobo will hold to ground as long as it senses that all voltages are OK. If something goes wrong (CPU cove voltage drops, for example), the "PS_OK" signal is released from ground (floating), and the PSU immediatly shuts down.
I studied the ATX spec that I downloaded from here:
http://www.intel.com/design/motherbd/atx.htm

Then I purchased a thermal switch from there:
www.MECI.com

I hoked the thing together, and it worked perfect (I tested it by applying the thermal switch against a soldering iron).
I came up with this protection AFTER I experience a meltdown. :-(

This is the simplest system. I consider it safer that software protection. If you set your core clock to close to the stability limit, then the conputer will crash, preventing any software protection to act. You can still back off to make sure it won't crash (back off ??? NO WAY ;-) ).

The best fail safe design, in my opinion, is a system that will read the temp of the CPU (I think the staff here has tested that) and will shut down the PSU (can be done with a relay on the green wire) when the CPU core temp exceeds a define threshold. The "Thermal switch" design only senses the water block. Better than nothing.

Hope this was helpfull.

[Since87]

I suggested the same thing as GigaFrog recently in oc-forums. Digikey carries a wide range of these switches too.

When it comes to protection, low tech is the way to go.

[bigben2k]

Originally posted by warpath
the problem with turning it off via MBM is that the speed at which readings are measured is far too slow to effectively shut-down before heat death occurs.

MBM is good for some cases, but useless if the cooling device (hsf, waterblock) falls off (which it won't, right?)

MBM may or may not work in all cases, since it's a software solution (it must be running, along with the OS). Also, there's an inefficiency factor in using processing power to monitor itself, but that's just a principle aspect.

The thermal switch looks like one of the best options, but it must be installed carefully.

Intel has a self-throttling device built right into the processor. AMD specs call for the mobo to immediately shut down.

Advantage, disadvantage. We need a throttling down solution for AMD, and it has to be a hardwire solution (i.e. a circuit).

[abyssling]

why not use waterblocks on your cpus and put another block in your loop directly after your rad, and a tec on the block to super cool the already somewhat cooled water going thru it, a small chiller, then put a aircooled hsf on the tec to cool the hot side of the tec (gotta swap the fans airflow to suck instead of blow)either venting it to an exhaust or mounting it externally...

wallah, simple chiller and very little condensation issues, if you want to get more complicated, add a temperature and humidity controlled tec controller to the setup to keep your temps above the dew point so your condensation worries are nil

[Since87]

Radiators and water chillers don't mix.

If the chiller is getting the water temp 1C below ambient, then the radiator will actually be heating the water up, rather than cooling it down.

[abyssling]

maybe i dont understand something, why would the radiator heat up water when the water goes thru rad before it goes thru the chiller?

[Since87]

A radiator exchanges heat between coolant and air.

If the coolant is warmer than the air, then heat travels from the coolant to the air.

If the coolant is colder than the air, then heat travels from the air to the coolant.

Edit: Maybe the piece you're missing, is that the waterblock only heats the water a fraction of a degree as it passes through.

[Arcturius]

Actually the biggest problem is that nobody has been able to make a successful water chiller from an inline water block (not that I've heard of...)

Look at Kevin's attempt from July of last year: http://www.procooling.com/articles/h...r_chille.shtml

[abyssling]

Lcpiper from pimprig managed to get his temps below ambient with exactly the setup i described. check it out HERE

Granted its not very far below ambient temps, but its a good start I think. with bigger/more powerful pelts I think it could work

see i understand how a radiator works, and i understand that the inclusion of a rad in your system will limit your lower ranges, but id think that with such a small chiller, you need to get your temps down a little in between the waterblocks heating the water and the chiller. seeing as how this chiller only cools a very small volume of water, surely it is beneficial to cool the water a little in the rad before it gets to the chiller?

can you explain why my reasoning is flawed?

*edit* nevermind i see the point about the block heating water only a fraction, hmm, so you are saying a chiller with no rad is more efficient?

[Since87]

Originally posted by abyssling
are saying a chiller with no rad is more efficient?

If the water temps are below ambient, they would be even lower without the rad.

If the water temp is above ambient the rad may be of some benefit, but the temp would need to be enough above ambient to justify the flow restriction that the rad presents to the water.

[abyssling]

do you know what total% of the water in the system or a set volume amount needs to be cooled in the chiller at any given time?

How do you determine the amount of volume to cool for your specific system?

Its obviously very easy to cool the water in a block, but cooling all the water in say a 20 oz reservoir would be somewhat more problematic.

[Since87]

I looked a little closer at Lcpiper's article. He's only using an 80 Watt TEC. He definitely needs the radiator.

As far as how much water:

That's very dependent on operating conditions and insulation.

See doc's post about 2/3 of the way down the first page of this oc-forums thread for information about a chiller system with a five gallon reservoir.

If you've got a smart controller system, a big reservoir can be an advantage because it allows you to "store cold" at low power consumption when the system is idle or off. Then when the system is loaded, that "strored cold" can be used along with the pelt chilling to maintain low temperatures.

With a dumb system (where you are just running the pelt at 12V all the time) with a continuously loaded CPU, you won't get much if any benefit from a reservoir.

[Arcturius]

Originally posted by abyssling
Lcpiper from pimprig managed to get his temps below ambient with exactly the setup i described. check it out HERE

Granted its not very far below ambient temps, but its a good start I think. with bigger/more powerful pelts I think it could work

But, his 'below ambient temps' are only 0.5C below ambient, and that's a water temp from the chiller to the CPU block. He is still getting CPU temps well above ambient.

On the other hand, it did get him 2C at the CPU, so that isn't bad.
Also in his favor: it is significantly more fault-tolerant than having the TEC on the CPU (TEC failure worst case: water is no longer chilled), but I personally don't think the gains are worth the power consumption. For those of us running standard water blocks, I think it makes more sense to run a big pelt (226W) or a pair of moderate sized pelts (120-156W) right on the CPU, and cool them with water.

But from another perspective, the inline chiller makes a lot of sense for those of us running direct-die.

So...I guess it really depends on how far you want to take it: really low temps with more severe consequences with failure, or small gains with relatively little risk. I suspect I know how the crowd around here will vote, though.