TEC1-12740 direct die thermal performance at various voltages (Skylake 6700K 4.8GHz)

cypres · Nov 22, 2015

Results of voltage/current adjustment on CPU core temp while running y-cruncher Benchmark Pi to 1,000,000,000 decimal places

Baseline overclock

Full results

12.5v 27.8a vs. 13.0v 28.8a

13.0v 28.8a vs. 13.5v 29.7a

13.5v 29.7a vs. 14.0v 30.8a

14.0v 30.8a vs. 14.5v 31.8a

14.5v 31.8a vs. 15.0v 32.4a

15.0v 32.4a vs. 16.0v 34.2a

cypres · Nov 22, 2015

I know the results aren't the clearest but the best performance seems to be in the 13.0v - 14.0v range. As expected when approaching vmax the parasitic losses/lower COP is quite evident. Here is an example of 13.0v compared to 16.0vmax

LiamG6 · Nov 23, 2015

just found your thread, I'm surprised there isn't more difference between all the voltages, I expected at least some pretty diff performance across the whole range but it's all virtually the same. what is the difference in max temp reached with all the voltages? ie what is max temp recorded in HWmonitor etc during the bench with all diff voltages? considerable difference? or all about the same? something is up. Have you removed the extra cold plate you added in? That will really be inhibiting the heat transfer and it is possible that you have too thick a piece of copper because you added in a second cold plate and your CPU is getting hotter than the cold side of the TEC due to slow heat transfer.

At the end of the day I think you don't have enough Qc from that TEC to handle your heatload. Still, its good to see some results. Curious to know if you removed that extra cold plate though.

Elloquin · Nov 23, 2015

My

The TEC is under torqued. That TEC in theory should have a sweet spot around 11 volts. I think he is chasing the parasitic losses which are exasperated by the lack of torque. I am assuming it's a max voltage 16v TEC he didn't say.

LiamG6 · Nov 23, 2015

yes it is 16v umax and yes something does appear to be amiss. may be torque as you say. i really have to get a torque screwdriver. any recommendations ello?

cypres · Nov 23, 2015

Quote:

Originally Posted by LiamG6

just found your thread, I'm surprised there isn't more difference between all the voltages, I expected at least some pretty diff performance across the whole range but it's all virtually the same. what is the difference in max temp reached with all the voltages? ie what is max temp recorded in HWmonitor etc during the bench with all diff voltages? considerable difference? or all about the same? something is up. Have you removed the extra cold plate you added in? That will really be inhibiting the heat transfer and it is possible that you have too thick a piece of copper because you added in a second cold plate and your CPU is getting hotter than the cold side of the TEC due to slow heat transfer.

At the end of the day I think you don't have enough Qc from that TEC to handle your heatload. Still, its good to see some results. Curious to know if you removed that extra cold plate though.

Max delta is really hard to say at this point, but it seems to be about 5*c when comparing 13 and 16v under similar conditions. But I agree, there is certainly something up with my TEC setup. One thing could be the paste, I'm using AS5 at the moment until my order of Gelid GC extreme comes in, but supposedly AS5 has a 200 hour burn in time and I don't think I've racked up more than an hour before running those tests. So that's one thing. Also I've been mounting/remounting the TEC and I'm fairly sure that the plates around the TEC are torqued plenty, but I don't have a torque driver to confirm. I've broken TECs before with a similar style clamping block so I think I can say that it's pretty darn tight. The block is clamped down with enough force to fully compress the springs and is difficult to torque any more by hand, and I've checked multiple times that everything is square. I've done a few different experiments with no additional cold plate and adding in plates of various sizes and I seem to get better results with the largest plate (50mm x 50mm x 10mm). I've also tried a 1/4" plate (what this was run on, seems slightly better than no additional plate) and a 1/8" plate (most negative impact). It could also be how I have the water block mounted - I have the barbs at the side with the input on the lower left and output on the upper right but I really need to rotate it 90 degrees to ensure the output is closest to the top and to push out any trapped air. While I don't think trapped air bubbles are an issue right now (I've taken it off numerous times and run the pump while rotating the block to free any) it still seems like a bottom to top push configuration would perform better than side to side.

I wish there was a more accurate way of measuring the true thermal output of the processor but here's where I'm at now. At lower voltages (1.32vcore) I can run prime95 with small FFTs (max heat) all day and never break 50-55*c, but higher voltages/clocks (>1.4vcore) and cpu temp will slowly raise past 80*+ and cause instability. It seems as if the TEC isn't able to move enough heat but for that to be true it would have to be generating >200W which I know is not the case. Actually now that I think about it my AXi can measure the power going to the motherboard connectors, I don't even think that's over 200w but I can confirm that.

cypres · Nov 23, 2015

Here's the datasheet for what it's worth: http://www.everredtronics.com/files/TEC1-12740_62x62x4.5mm.pdf

Yes, the TEC is 16v umax

I'm running my low voltage overclock now (1.32vcore) and total cpu power seems to cap around 120 watts as reported by ASUS AI Suite, and confirmed by using CorsairLINK to watch the current draw of the 12v aux and main connectors. I'll try it again with the baseline configuration later tonight if I have some time

Elloquin · Nov 23, 2015

@Liam I don't have any cheap ideas for a torque screwdriver. I got a good deal off amazon one day.
@Cypres Problem one is AS5. It doesn't work well cold. Works great hot just not the other way around. Also if you cracked a TEC with hand force either it or what was being clamped to it wasn't flat. Tec's need between 150 and 300 pounds of torque on them. You can't do it by hand. When you use a torque screwdriver and get to the set point you can actually hear the bolt make a "ping" as the wrench snaps.

Puck · Nov 23, 2015

Thanks for sharing.

I tried Thermene as a TIM for TECs and was initially very impressed since it was way better Temps then AS Ceramique2. Very thin paste with great spreading to form a super thing layer on lapped surfaces.

Unfortunately, it's lifespan under a TEC is terrible and the constant heating and cooling dries it out quickly, costing a lot of performance. Now I'm back to my Ceramique2, since it's the only thing that lasts besides CLP...which I can't use with my current aluminum cold plate.

My chip died unfortunately, but I have a 5ghz capable replacement on the way.

cypres · Nov 23, 2015

Quote:

Originally Posted by Elloquin

@Liam I don't have any cheap ideas for a torque screwdriver. I got a good deal off amazon one day.
@Cypres Problem one is AS5. It doesn't work well cold. Works great hot just not the other way around. Also if you cracked a TEC with hand force either it or what was being clamped to it wasn't flat. Tec's need between 150 and 300 pounds of torque on them. You can't do it by hand. When you use a torque screwdriver and get to the set point you can actually hear the bolt make a "ping" as the wrench snaps.

I suppose you could be right - it was a 40mm that cracked on my test bench and I only had 2 points of pressure holding it on to a quite large aluminum heatsink (scavenged from a fridge). But wow, 150-300 pounds, that's insane... the problem I see now is that the top of my waterblock is Delrin/POM and I don't want it to crack, plus I'm not even sure those screws could hold that amount of force

This is the block I'm using
http://thermo-electric-cooling.com/index.php?main_page=product_info&cPath=1&products_id=19

Also I removed the shim I was using and went back to just the cold plate that came with the TEC and trimmed the foam a bit more to ensure it wasn't obstructing the block. I also had enough ceramique to use for the cpu this time (I was already using it on the cold side of the TEC, AS5 on the hot side). Before I did all that I ran P95 small FFT for about an hour and was 60*c stable, and I'm about 20 minutes into the second run and temps are the same, but I did notice the initial "spike" in temps is a lot smoother. Still doesn't seem right since the CPU is only putting out ~120w and water temps are right at 40*c.

cypres · Nov 23, 2015

Quote:

Originally Posted by Puck

Thanks for sharing.

I tried Thermene as a TIM for TECs and was initially very impressed since it was way better Temps then AS Ceramique2. Very thin paste with great spreading to form a super thing layer on lapped surfaces.

Unfortunately, it's lifespan under a TEC is terrible and the constant heating and cooling dries it out quickly, costing a lot of performance. Now I'm back to my Ceramique2, since it's the only thing that lasts besides CLP...which I can't use with my current aluminum cold plate.

My chip died unfortunately, but I have a 5ghz capable replacement on the way.

Sorry to hear about your chip... this is my 2nd Skylake for what it's worth

And thanks for the warning about Thermene, I'll stay away from that

Elloquin · Nov 23, 2015

LOL

Not the mounting screws and springs. The screws that hold the TEC between the hot and cold plates. They are independent. Once you torque the TEC you mount it like any other water block. As for tim just use mx-2 cheap and reliable.

cypres · Nov 23, 2015

Quote:

Originally Posted by Elloquin

LOL

Not the mounting screws and springs. The screws that hold the TEC between the hot and cold plates. They are independent.

Hahah yeah I know

I mean the head of the screw that tightens the TEC still exerts force against the plasticy bit, kinda like this:

(screw head)=(cover/no threads)=(cu water block/no threads)=(TEC)=(cold plate, threaded)

So every time I read your comment I reach in my case with a allen wrench and pliers and torque it a little more, and the funny thing is I noticed the current drops as I tighten it down. It's just about to point where I think I need to get the right tool to make sure they are tightened evenly though. But they are pretty darn tight.

LiamG6 · Nov 24, 2015

how thick is the cold plate that comes with the block cypres? I'd avoid all extra cold plates unless you are actually just swapping out the original cold plate for a thicker/thinner one, there is a sweet spot to be found with cold plate thickness to handle loads spikes and long duration load. as ello states, torque it down correctly, and get a torque wrench/screwdriver. 2 mount bolts on opposite sides of TEC would have actually been better, it is what is recommended from the TEC retailers, 4 on that block is fine though and will reduce the torque/screw.

300ft/pounds of torgue ello? beauty I can use my torque wrench haha. Problem is I don't have a cold plate that bolts to the block, I just mount it all to mobo and tighten it down, aint no way I'm putting that much torque on my mobo ahahahhahahaha, I'll wait till I buy the 62mm block. I'll have to do the equation for the torque required.

I'm using thermal grizzly kryonaut TIM, thick and hard to apply, like gelid gc extreme, but good on hot and cold sides, about 5*c better than Arctic Ceramic 2

hot side water temp of 40*c is a little concerning, be better if that was 25-30*c. the hot side of the TEC will actually be slightly hotter than that 40*c. I'll check the data sheet of TEC to see what i can find. what is your ambient temp?

LiamG6 · Nov 24, 2015

ps, you need to get a thermistor/thermocouple drilled into that cold plate to see what temp your cold plate is holding, or even just a flat temp sensor with an LCD read out from XSPC stuck to the bottom of cold plate right next to CPU IHS, you really need to know what is happening at the cold plate. ie if the cold plate temp is holding steady at load or slowly rising in temp etc.

plus I need to know what sort of deltas you are getting across the TEC, you said hot side water is 40*c, so say hot side of TEC is like ~45*c, may be hotter as you have so much wattage going through there, if the *c/w loss of that block is like 0.012*c/w and you have 600w going through it thats a 7.2*c temperature differential from your water temp, so if water is 40*c then hot side of TEC could be 47.2*c, thats why it is so important to keep your total wattage down as the *c/w losses really kills your delta. the 669w TEC is much more efficient than your 12740. you need a temp sensor on that cold plate to find out what is going on. you may just be pumping way too much heat through that block.

what we need to know is as much temp details as possible from your CPU, hot side water, TEC hot side and cold side etc with CPU in idle and at load with an approximate CPU power consumption figure from your software as well as the voltage and amps of your TEC. you really need to be able to monitor all these to make sure your setup is performing optimally. mainly that TEC cold side and hot side delta is the important thing to know

cypres · Nov 24, 2015

Quote:

Originally Posted by LiamG6

ps, you need to get a thermistor/thermocouple drilled into that cold plate to see what temp your cold plate is holding, or even just a flat temp sensor with an LCD read out from XSPC etc, you really need to know what is happening at the cold plate. ie if the cold plate temp is holding steady at load or slowly rising in temp etc.

plz get out of my head

I had the same idea... and the results are kinda... well... spooky

So any idea why my core is 80*c higher than my cold plate?

edit: fyi the red lcd thermocouple is on the waterblock/hot side that the TEC is touching and the blue is on the bottom of the cold side, on the stepped portion. The thermocouples are just stuck on with clear tape and a tiny bit of ceramique. I swapped the red and blue to make sure the screen isn't broken with same results. I can see the cold side getting warmer but it doesn't break -40*c

cypres · Nov 24, 2015

Quote:

Originally Posted by LiamG6

how thick is the cold plate that comes with the block cypres? I'd avoid all extra cold plates unless you are actually just swapping out the original cold plate for a thicker/thinner one, there is a sweet spot to be found with cold plate thickness to handle loads spikes and long duration load. as ello states, torque it down correctly, and get a torque wrench/screwdriver. 2 mount bolts on opposite sides of TEC would have actually been better, it is what is recommended from the TEC retailers, 4 on that block is fine though and will reduce the torque/screw.

300ft/pounds of torgue ello? beauty I can use my torque wrench haha. Problem is I don't have a cold plate that bolts to the block, I just mount it all to mobo and tighten it down, aint no way I'm putting that much torque on my mobo ahahahhahahaha, I'll wait till I buy the 62mm block. I'll have to do the equation for the torque required.

I'm using thermal grizzly kryonaut TIM, thick and hard to apply, like gelid gc extreme, but good on hot and cold sides, about 5*c better than Arctic Ceramic 2

hot side water temp of 40*c is a little concerning, be better if that was 25-30*c. the hot side of the TEC will actually be slightly hotter than that 40*c. I'll check the data sheet of TEC to see what i can find. what is your ambient temp?

It looks about 1/4". From what I read the ideal is around 3/8ths but that was an old article with probably lower powered TECs and processors. I was just adding copper between the cold plate and CPU, I don't have anything to tap holes the correct size at the moment, so that's out.

Ambient stays about 22.5*c, but with just the one triple radiator I'm actually very, very happy with 40*c. Great efficiency if nothing else.

And yeah don't put 300ft/lbs on the chip lol that poor socket will probably explode XD

LiamG6 · Nov 24, 2015

have you delidded your CPU and applied Coolaboratory Liquid ultra or pro to the die? also I would encourage you to lap your IHS, all my IHS on haswell have been extremely concave and I dropped 20*c on my core temps just with adding CLU TIM to the die and lapping the IHS.

that -49*c is clearly wrong, that is an 90*c delta which is not physically possible with a single TEC

at minus temps I believe those temp sensors lose accuracy, might need a thermistor or a k type thermocouple

its good that 39*c is your hot side block temp though and not your water temp, so water is about 30*c most likely.

yeah I'd break the board well before 300ft/lbs hahaha

cypres · Nov 24, 2015

Quote:

Originally Posted by LiamG6

have you delidded your CPU and applied Coolaboratory Liquid ultra or pro to the die? also I would encourage you to lap your IHS, all my IHS on haswell have been extremely concave and I dropped 20*c on my core temps just with adding CLU TIM to the die and lapping the IHS.

that -49*c is clearly wrong, that is an 90*c delta which is not physically possible with a single TEC

at minus temps I believe those temp sensors lose accuracy, might need a thermistor or a k type thermocouple

Yeah I think you are right about it not being accurate but it's nice to dream. Not delided or lapped, but I think you are right about it being concave. When I was applying paste this last time I could notice it that it was spreading very thin in a pattern around the edges. I guess that will be my next project... not sure if I'm going to delid or not yet though. Really don't want to risk it.

LiamG6 · Nov 24, 2015

you should look into de8auers delid die mate, its a tool for safe and foolproof delidding/relidding, looks fantastic and shouldn't be that expensive, can be used for Ivy/Haswell/Skylake and most likely future mainstream chips.

my method to delid is to clamp CPU and IHS in a clamp, one side on substrate, the other side on IHS, apply a slight amount of tension, then get a hair dryer and heat the IHS, increase the clamp tension a bit then keep going with that method until it "pops" ie glue will release the IHS and it will pop over a mm or two and then just peal the IHS off. less likely to damage anything than with razor/hammer/or plain vice method, but not as good as de8auers tool.

you really need to delid and lap to get a better result

the fact that your cold side temp only rises 9*c at load is interesting, even if actual temp is wrong the delta between load and idle should be right, so say you might be at an actual 0-10*c cold side temp. or -10*c to 0*c at best

What is your CPU core temp with CPU in idle?

TEC1-12740 direct die thermal performance at various voltages (Skylake 6700K 4.8GHz)

cypres

cypres

LiamG6

Elloquin

LiamG6

cypres

cypres

Elloquin

Puck

cypres

cypres

Elloquin

cypres

LiamG6

LiamG6

cypres

cypres

LiamG6

cypres

LiamG6

Top Contributors this Month

Recommended Communities