Peltier woes.

[Muggy]

I've spent 2 days trying to get my PelTEC cooler to work with my Slot 1/GlobalWin VOS32 combo. I ordered an 85 watt kit and did everything necessary to mount it to my heatsink. It wasn't easy but it appeared to be tight and sturdy.

So why are my temps a good 3 or 4 degrees higher? I get max temps of 42C with my normal HSF solution. I didn't see much of a change with my pelt setup. I even ran it switched on and off to see if there was a change. I didn't see any difference!

The strange thing is that I verified the setup prior to installing my CPU. It frosted over and the heatsink got a bit warm. Just like it's supposed to. When it was running, I managed to stuff a thermal probe onto the coldplate and measured about 19C. Considering the heatsink temp was about 40C, I figured that was about right. My CPU was registering about 2 degrees higher though!

Am I missing something? Here's the specifics- 85 watt peltier, 3/8" copper coldplate, tons of insulation. All of this is screwed together with plenty of pressure. I tried clamping the coldplate to the heatsink, but there wasn't enough clearance to get the CPU to fit. I need some suggestions before I decide there's something wrong with the kit and send it back.

[Naeleros]

I'm not peltier expert...

But, possibly your HSF is not sufficient to remove all the heat from your Peltier + your CPU? If that would be the case.. it would actually heat up over the temp of just the HSF + CPU. (Because the Peltier heat would just be extra...).

I would double check to see if there's a way to increase contact between the HSF and the Peltier... the Peltier and Coldplate... and the Coldplate and CPU.

Of course... make sure you wired it properly. Some people have wired it to the improper voltage and gotten poor results.

James

[dimmreaper]

Is the coldplate lapped? It's standard operating proceedure to lap both sides of the coldplate.

Is the heatsink lapped? It is also standard operating proceedure with peltiers to lap the heatsink/waterblock to insure a good contact.

Are you using thermal compound? Don't forget to use it at every thermal junktion (IE both sides of the peltier, as well as on top of the CPU die). ArcticSilver or, comparable silver compound, is highly recomended whenever you have a high temperature differential situation (IE peltiers).

[dozier768]

seems like if its frosting the colplate its moving plenty of heat, perhaps the thermal probe is way off

[dimmreaper]

Originally posted by dozier768
seems like if its frosting the colplate its moving plenty of heat, perhaps the thermal probe is way off Honestly, you don't have to move plenty of heat to get the coldplate to frost over with no load. You don't even need thermal compound for that.

But I $#|T you not, temps on a CPU with a peltier will be much lower with lapping and the use of good thermal paste.

With just a CPU and heatsink you have only one thermal junction. With a CPU,heatsink/waterblock,peltier, and coldplate you now have 3 thermal junctions. If you lose just 5% of your cooling ability per junction, you have lost a lot with three junctions!

[dimmreaper]

Just an Idea Muggy. But maybe you should screw studs in to the heatsink, and use a wider coldplate with fourholes in it. Then you could squeeze the peltier supper tight between the heatsink and coldplate with a couple of thin nuts. Then attatch the CPU to this (the studs need to be long enough to pass theough the SECC package), and snug it up with four more nuts.

[Muggy]

Originally posted by dimmreaper
Just an Idea Muggy. But maybe you should screw studs in to the heatsink, and use a wider coldplate with fourholes in it. Then you could squeeze the peltier supper tight between the heatsink and coldplate with a couple of thin nuts. Then attatch the CPU to this (the studs need to be long enough to pass theough the SECC package), and snug it up with four more nuts.

I tried that with some brass straps I made. I had 4 bolts running through the HS and then the peltier/coldplate strapped with nuts holding it all down. The trouble is, the gap between the CPU die and the rest of the proc isn't large enough to allow the straps. I ended up preventing the die from contacting the coldplate. Not good. I could use a thinner strap I suppose, they are readily available. I was a bit worried about heat transfer back to the coldplate with this method though, since the bolts will conduct heat back through the straps onto the coldplate.

The larger coldplate would be perfect, except that there isn't space to fit nuts between the CPU and plate. I suppose I cold drill the corners of the colplate out and solder nuts in...

I'm still stumped by the cold temps at the edge of the coldplate though. I thought the purpose of the coldplate was to disperse heat across the entire surface of the peltier. Why then was the core reading 40+ and the outside of the coldplate 19? I would think the entire coldplate would be more consistent. Any clues?

Also, to answer one of your first questions, dimmreaper. I did in fact lap all contact surfaces except the peltier (kinda hard to lap ceramic). They are all very smooth and reflect quite nicely. It was almost a shame to have to cover them up. I'm using standard RadioShack thermal grease on everything and haven't heard of anyone who got terrible results doing this.

I've read up on about every peltier project available on the net and mine still seems different. The Leufken's kits I've read about use a 52watt peltier and they achieve negative temps idle and a few degrees over ambient at full load. This is with assemblies that don't have a clamp on the peltier/coldplate. They also run the tests on processors usually 600Mhz or higher and they tend to be overclocked in the mid to upper 900's. The temps I was getting were at the standard 700Mhz setting with default voltage!

Keep the suggestions coming guys! I appreciate 'em.

[dimmreaper]

Is the 40C reading coming from an on-die temp diod? What MoBo are you using? Are your 40C temps coming from the BIOS or are you using a program like MotherBoardMonitor (don't try reading On-Die temps with a VIA southbridge using MBM, it dosn't work).

Are you sure you have the TEC wired up to 12V? You need to be sure the TECs red lead is wired the the PSUs yellow leads.

[Muggy]

You can check my sig for specs, but here it just in case:
Abit BX6-r2 board using Intel 440BX chipset.
Thermal diode temps are used for CPU temp
External thermister that is accurate +/_ 1% for other measurements.

I'm using an external 12V 20amp powersupply for CB gear. It puts out about 13.8V under full load. I've checked and verified this using a very expensive Fluke multi-meter. Any other thoughts?

[dimmreaper]

Well since your 42C+ reading is an on-die diod on a board that supports it, It must be accurate.

I don't know, maybe your just using too much thermal compound, It's been done in the past. Your using just a thin layer right?

I wonder if your coldplate could be too think? Every type of matter has some thermal resistence, perhaps your coldplare is thicker than optimal. Most often we use 1/32" to 1/4" thick plates with 1/16" and 3/32" being used most widely. 3/8" seems a bit excesive. Do you have any 1/16 aluminum around to give a crack at it with?

[Muggy]

I considered that as well but thought that a thicker coldplate would actually help since it offers more metal to spread out the heat.

[Colin]

Originally posted by Muggy


I'm still stumped by the cold temps at the edge of the coldplate though. I thought the purpose of the coldplate was to disperse heat across the entire surface of the peltier. Why then was the core reading 40+ and the outside of the coldplate 19? I would think the entire coldplate would be more consistent. Any clues?


The temps of the coldplate near the CPU core will be near your CPU temps. It's common to have colder temps away from the core. Think about the heat path. I get about a 20C difference between coldplate and CPU temps in my dual 120 watt pelt Athlon system.

[Muggy]

Originally posted by Colin

The temps of the coldplate near the CPU core will be near your CPU temps. It's common to have colder temps away from the core. Think about the heat path. I get about a 20C difference between coldplate and CPU temps in my dual 120 watt pelt Athlon system.

So what kind of temps are you getting? By the way, Don informed me that I got a 72 watt pelt, not the 85 watt version I requested. Too bad that's the one he priced me. I'd be a bit more upset if it weren't for the fact that he was sending it for the same price. Still... Oh, well.

I still think my temps aught to be lower than what they're showing. Also, I failed to mention that with the peltier switched off, I didn't notice any change in temps. This seems very strange indeed. Keep the comments coming!

[dimmreaper]

Originally posted by Muggy
Also, I failed to mention that with the peltier switched off, I didn't notice any change in temps.Oh! In that case it soundls like one of the following:

a. Your not reading the right temp source with MBM (or whatever temp reading prog you are using). It sounds to me like maybe your reading the chassis temp.

b. 19C isn't even that cold, so maybe the peltier is not working properly. In which case send it back, and maybe the bigger one while your at it.

c. The retailer sent you a 54W peltier by mistake.

[Muggy]

I was worried that might be the case! I also had another thought. Is it possible that the cpu not being fully seated in the slot would through off the thermal diode monitoring? I noticed my home grown backplate was touching the slot. I wasn't able to check how much farther it would go in since the backplate is too rigid (it's made of lexan).

My CPU temps are from the internal diode, no doubt about that. I've only got 2 sensors being monitored in MBM and they haven't changed at all. So what I'm seeing is at least consistent between active peltier cooling and standard air cooling. Whether or not they're accurate really doesn't matter in that sense.

I should get an email back from PelTEC today. We'll see if he's got any suggestions. I think I'm going to try my strap technique again. Only this time with thinner straps and possibly isolate the bolts with plastic on the heatsink side. I'll let everyone know what I find out.

[ve6jhc]

Try running your Peltier cell at 5 volts. I have tried my cells at voltage from 1-16....Anything after about 9 volts seems to produce extra heat but no more cooling.....

[Les56]

A few thoughts
1) The temp gradientof ~20c is consistent with Colin's experiences and indicates normal Peltier coling is occuring.
2) Do you have a poor electical connection to the Peltier,ie is the the current correct for the applied voltage.Using a 172watt Peltier I found that the current was nearly 2amp below that predicted by Kryotherm( http://www.kryotherm.spb.ru/soft.htm ).I disccovered that the Molex connector i was using was very warm and replacing with a soldered joint solved the problem.
3) I think ve6jhc's suggestion is sound.I also have run a 85 watt peltier at 1-16v and have found the best cooling could be found at 9v in some circumstances.Cooling a Celeron @1000ish and 2,2v(~50watt) with 3 different waterblocks I found that with the worst block the best cooling was at ~9v and with the best block the best cooling was at the max voltage of 16v.With air cooling this optimum voltage could well have been lower.

[Muggy]

Originally posted by ve6jhc
Try running your Peltier cell at 5 volts. I have tried my cells at voltage from 1-16....Anything after about 9 volts seems to produce extra heat but no more cooling.....
I don't have any way to vary the voltage to the peltier. I'm stuck at 13.8 since this is a high current 12v bench power supply and 13.8 is about the industry standard. Also this peltier is supposedly a 72watt unit, not the 85watt version I was told I'd be getting.

[Muggy]

Still no word from PelTEC. Any other ideas?

[dimmreaper]

Use your digital multi-meter to test the PSUs voltage while under load (connect the meters leades in parallel with the peltier).

19C doesn't seem right, that is like two C below room temperature. You should be seeing closer to 10C, I would think. So it either has to be a bad pelt, or something wrong with the PSU.

[Muggy]

Originally posted by dimmreaper
Use your digital multi-meter to test the PSUs voltage while under load (connect the meters leades in parallel with the peltier).

19C doesn't seem right, that is like two C below room temperature. You should be seeing closer to 10C, I would think. So it either has to be a bad pelt, or something wrong with the PSU.
I did that the first time I fired it up. I had an old 230 AT PSU laying around and the pelt dropped the 12 volt leg to 11. That's when I brought in the 20 amp supply. It's reading 13.8 under load and 13.8 without a load. Just what a properly functioning regulated power supply should read. The heatsink fins on the PSU are getting nice and toasty as well.

I did some figuring with the excel spreadsheet file from this site, called OverCPU. Even at worst case scenarios (hot side temps of 60C, lower tDelta, 72watts) it shows my CPU temps at around 15C. I cant imagine this setup being this bad unless there's a problem with the peltier. Oh, and I'm still waiting to hear from PelTEC. I was kinda getting to like them as a company before I placed my order. I haven't written them off yet though. Stay tuned...

[Muggy]

Still no word from PelTEC. And a friendly "BUMP.":)

[dimmreaper]

Muggy, PelTEC sponsors the site for which I am now Editor, www.wildandyc.co.uk , would you like me to contact Don on your behalf and see if I can get you some help from PelTEC on this?

[Muggy]

Originally posted by dimmreaper
Muggy, PelTEC sponsors the site for which I am now Editor, www.wildandyc.co.uk , would you like me to contact Don on your behalf and see if I can get you some help from PelTEC on this?
Thanks Jeff! I'd love to take you up on that but I'll give Don a bit more time to respond first. I emailed him on Monday so I'll give it till tomorrow for a reply. He's been really helpful up to this point. I'm sure if there's a problem, he'll try to make it right. I will let you know what I find out. And whether or not I'll need "The Big Guns!" By the way, wildandyc's site rocks! I've been checking his pages out for some time now. Glad to know you're on my side.

[dimmreaper]

Originally posted by Muggy
By the way, wildandyc's site rocks! I've been checking his pages out for some time now.You ain't seen nothin yet, keep an eye on us. We have some more cooling stuff coming out of the "skunk-works" in the coming months. Sorry I can't comment further, but I've said too much already . . . .

[Muggy]

Still no word from PelTEC... Might be time to call in the 'reaper.' Jeff, if the offer's still good, I'll send you all the particulars of my order and such if you'd like.

[dimmreaper]

E-mail me the particulars, and I'll forward 'em with my blessing to PelTEC.

[Muggy]

Case closed on this one! Turns out my BX6-r2 suffers from something I hadn't thought to check. No thermal diode support! I'll have to mod my board and then I'll be back to post result.

I have to admit I feel pretty stupid about this. I'd even read the articles about enabling the internal P3 diode support and somehow thought my board didn't have that problem. I thought the default sensors in bios referred to the internal cpu temp and the external sensor. The BX6.2 has 3 thermal sensing inputs to its winbond chip: A system temp, (of which I know not the location for the sensor) an external sensing input (common 10K ohm thermister works great) and the CPU's internal thermal diode. My board is one of many that doesn't have the internal diode sensor enabled. I so wish I would have known this before! My sincere apologies to all who attempted to help me on this. I should have known better. More details later...