Omega Thermocouple (in IHS) Vs. Onboard Sensors

[MaxxxRacer]

nik, showing the difference between Tcase and Tdie is not uselss, but rather the data provided from ITE is rather useless as the accuracy of it is dubious at best.

with the maxim kit hooked up to the cpu after calibration, showing both Tcase and Tdie would be VERY interesting..

Plywood99, i showed that the delta is not consitent with different ambients.. nik was wondering aobut consistency between different waterblocks.

[snowwie]

How could having both sets of data hurt?

how will they help? they will not in any form invalidate or validate one method or another, they will do nothing but raise questions (mostly old) that cannot be answered by anyone here. please step back and look at what maxxx is trying to do here: reliably measure the temperature as relates to thermal tranfer. and I still don't get how the DIE TEMP you are incessantly preaching about will get him closer to his goal.

[MaxxxRacer]

I will also be calculating heat output by measuring the dT between waterblock in and waterlbock out temperatures... along with flow this will give wattages... with only .1C resoution im not sure how useful it will be, but it will give an approximation.

[nikhsub1]

please step back and look at what maxxx is trying to do here: reliably measure the temperature as relates to thermal tranfer. and I still don't get how the DIE TEMP you are incessantly preaching about will get him closer to his goal.

Only measuring Tcase IMO is not reliable. I you don't get how the die temp relates to anything, put your head back in the sand.

[snowwie]

okay, i see we all have opinions...

you say Tcase is not reliable
maxxx says Tdiode is not reliable

i'm glad we've made so much progress (maxxx what are you getting yourself into?!)

[MaxxxRacer]

snowwie, I ask myself that on a regular basis..

to make everyone happy, the best way is to measure Tdiode with the maxim kit as well as Tcase with the embedded thermocouple (although an RTD probe attached to a high end fluke or omega .01C resolution thermometer would be better)... but I refuse to muck with this CPU more than it already is for fear of breaking it.

[aicjofs]

There is no doubt that every motherboards reading of the on die sensor is questionable from board to board. But 8C here doesn't seem all that crazy to me though based on silicon and AMD TIM thermal resitivity. The only good way to show this effect is to hold the same IHS probe temp constant and record say 10 different boards on die reading. Then we could actually say the "motherboard sensors" are very inaccurate.

My thoughts on this "on die" vs "probe temp" data... If we look at it from the perspective of testing water blocks only then the on die temp doesn't really matter because testing on the IHS is sufficent. However for completeness there is no reason not take the on die temp reading as it has side interest for learning other things besides how a water block performs(i.e. a better understanding of IHS temp vs on die temp.

On a side note: I was just thinking to myself about the calibration methods derek used on the 1700 and A64 chip... they were done with no power going through the chip... what happens when you put power through the chip? does the calibration still hold true, or is it effected by the various signals in te chip?

I ask this question not to attack Derek or his methods (infact I tried the same method and learn a good portion of what I know about testing directly from him), but rather as an honest question.

If the on die snesor is effected by the chip running or not running, that does indeed pose an interesting challenge. and on the same note.. Is the thermocouuple I have embeded in the IHS affected by the chip running too? and if so, how would we ever go about overcoming such obstacles?

Well we can safely say if you are generating any heat(power applied) in the core and calibrating to an extrenal source then all bets are off, and why the calibration should be done with no power. Does signals or EMI have an effect? Now that's a good question. I'll say if you are always running the chip and the same freq and voltage you could say that any signal interference would at least induce the same .1C (or whatever it might be) shift. If you are just testing water blocks I don't think it would matter because each block is getting the same shift.

Will there ALWAYS be an 8C difference between core and IHS? Will that gap widen or shorten depending on the block? All very interesting and not useless at all.

Of course the 8C will change. All factors held constant it depends on CPU heat load. I think we have already seen a glimpse of this by Maxxx's claim of a small delta at idle, exactly what I would think would happen. If we start changing factors, in this case the water block, then we will probably see a difference in the delta from the best to worst block. I agree with you as I said above that the data should be taken, but it would be for a different study then the study of water block performance. You only need one data point on the "other side" of the water block to determine performance.

If you could put 1,000 probes from the center of the die out to the center of the water the temperature at each probe would look something like this with temp on vertical and distance from core on the horizontal(except at turbulent flow it isn't linear but I'm too tired to draw it right also assumes constant water temp). This isn't to scale or anything, but the point is the steeper slope at load, then at idle, and at off

[Plywood99]

"My thoughts on this "on die" vs "probe temp" data... If we look at it from the perspective of testing water blocks only then the on die temp doesn't really matter because testing on the IHS is sufficent. However for completeness there is no reason not take the on die temp reading as it has side interest for learning other things besides how a water block performs(i.e. a better understanding of IHS temp vs on die temp."

I am very interested in these temperatures. For instance I and many others are convinced that the tim1 layer on AMD cpu's "degrades" over repeated mountings. Recording core and ihs temps over many many mountings will prove or disprove this.

[Mr. Tinker]

I just can't trust this experiment since the type of sensor is not the only variable.

I can easily believe that two regions of the die at one time could hold temperature difference of 8C or more. MOst people imagine the core to be one uniform temp, but it is not. 0-dimensional data describing a 3-dimensional subject.

[Bloody_Sorcerer]

*sigh* i'm getting sick of this discussion. will it make that big of a difference in the final results? i doubt it. theres always going to be the minutiae and so on and so forth... deal with it. I still dont understand why people have been moving away from heat dies and on to actual processors. fully loading a processor properly consistently would be a lot harder than just flipping a switch in my opinion. The whole testing methodology debate/debacle is quickly spiraling out of control.

[WeStSiDePLaYa]

i wouldnt call this completely accurate, your measuring OUTSIDE of the core, and through a piece of metal which is dissipating heat ALL around the IHS. your sensor isnt getting the full concentration of the heat.

and its no where close to being inside the core.

[MaxxxRacer]

aicjofs: I like that graph.. what software did you make it in? Im looking for good graph making software. (for reviews)

i wouldnt call this completely accurate, your measuring OUTSIDE of the core, and through a piece of metal which is dissipating heat ALL around the IHS. your sensor isnt getting the full concentration of the heat.

I make no claim of being completely accurate. If I were to make such a claim, all of you would be justified in slapping me with a 20lb trout.

While the IHS is disipating heat ALL around itself, you need to take not that MOST of the heat is being disipated directly above the core and not near the extremities.. and infact, the more effeciently you cool the core, the less heat gets transfered towards the edges of the IHS.. this has been documented by Joe at overclockers.

[FLB]

While the IHS is disipating heat ALL around itself, you need to take not that MOST of the heat is being disipated directly above the core and not near the extremities

I disagree. If you look at robotech's test results you'll see that the edge of the heat spreader is ~4.5C above the water temp. When you take into consideration that the ihs is in contact with a efficient waterblock and more than 9 times the size of the core, this 4.5C has to be considered huge. Such temps for such a large area can only be reached if alot of heat is passing through it.

In case of a block like the apogee, I would wager that as much as 80% of the total heat load is being transfered outside the primeter of the core

The next question is why is so much heat being spread through a relativly thin IHS. "My theory" is that it has to do with the high resistance (compared to the copper spreader) of the tim joint. As thin as the spreader is it's much easier for the heat to move lateraly than to move straight through the TIM joint. The more the heat spreads, the lower the TIM resistance becomes (due to the now larger heat transfer area).

Block design also plays a role and I think a block like the apogee compliments the above effects.

[MaxxxRacer]

In case of a block like the apogee, I would wager that as much as 80% of the total heat load is being transfered outside the primeter of the core

That defys all known ideas of thermodynamics. If you look at a thermal image of the heat transfer you will see most of it going straight up. A bad TIM1 joint will not make the heat go everywhere but up.. The heat still has to enter the IHS through the TIM1 joint.

I think I will have to thermal epoxy another probe to the side of the IHS to show some more data, and show the delta between IHS center and IHS edge for different waterblocks.. Now only if only I had a 10 input thermometer.

[FLB]

That defys all known ideas of thermodynamics.

All the better! For too long people have thought that the IHS had minimal effect. I've given it a lot of thought myself and the above conclusion is the only one that can explain the high edge temp that robotech observed. I'm looking forward to your results

[MaxxxRacer]

Well to be fank a 5C delta between water and IHS edge is pretty small. It is by NO means insignificant, but compared to the delta between core temp and water, there is a much larger difference.

[FLB]

Well to be fank a 5C delta between water and IHS edge is pretty small. It is by NO means insignificant, but compared to the delta between core temp and water, there is a much larger difference.

You need to take the size into proper account. For instance, at half the temp of the IHS center, the 9x larger outer perimeter will still be transfering ~4.5 times the heat load of the core area. This is assuming that the transfer efficiency of the block is the same throughout the contact area, which of course it isn't.

So the next question is how much does it differ. I don't expect a block like the apogee to differ much (my guess, 20% tops)

Question regarding your test setup: What speed and volts is the core running at? Does the water temp and IHS temp report the same thing when the computer is off (pump still running of course)? I'm only asking this because the 5C temp at idle seems a little high to me.

[MaxxxRacer]

I will most likely be running at 2.3ghz with 1.600V going through it. Rather high voltage, but the higher heat output helps with measurements. bigger deltas are alot easier to measure.

This is assuming that the transfer efficiency of the block is the same throughout the contact area, which of course it isn't.

well that makes your entire point (that paragraph) rather pointless..

[FLB]

well that makes your entire point (that paragraph) rather pointless..

Hardly. Even if we make an assumption that the heat transfer efficiency is twice as good over the core as it is over the rest of the surface, we still end up with 2.25x more heat being transfered outside the core.

So with half the surface temp and half the block efficiency, ~70% of the heat load will still be transfered outside the core area and there is no way that a block like the apogee will differ that much.

[Plywood99]

I disagree. If you look at robotech's test results you'll see that the edge of the heat spreader is ~4.5C above the water temp. When you take into consideration that the ihs is in contact with a efficient waterblock and more than 9 times the size of the core, this 4.5C has to be considered huge. Such temps for such a large area can only be reached if alot of heat is passing through it.

In case of a block like the apogee, I would wager that as much as 80% of the total heat load is being transfered outside the primeter of the core

The next question is why is so much heat being spread through a relativly thin IHS. "My theory" is that it has to do with the high resistance (compared to the copper spreader) of the tim joint. As thin as the spreader is it's much easier for the heat to move lateraly than to move straight through the TIM joint. The more the heat spreads, the lower the TIM resistance becomes (due to the now larger heat transfer area).

Block design also plays a role and I think a block like the apogee compliments the above effects.

FLB, you are forgetting a couple things here.

1) Maxx is correct when he said most of the heat is transfered directly above the core. But there are other things at work here. The ihs is also covering the whole chip. So heat from the sides of the core and from the surrounding silicon will be transferd to the ihs through the surrounding air and the ihs edge contacting the chip itself. This is what is causing the majority of the temperature difference.

2) Your thoughts on tim1 are totally incorrect. Since tim1 is the only thing connecting the core to the ihs, almost all the heat must pass through tim1. There is no lateral spreading happening the way you mentioned it. Remember tim1 covers the core and there for is basically the same size as the core.

[FLB]

2) Your thoughts on tim1 are totally incorrect. Since tim1 is the only thing connecting the core to the ihs, almost all the heat must pass through tim1. There is no lateral spreading happening the way you mentioned it. Remember tim1 covers the core and there for is basically the same size as the core.

I wasn't talking about TIM1. I was talking about TIM2.

[Plywood99]

I wasn't talking about TIM1. I was talking about TIM2.

Lol, so you were. My brain is on autopilot sometimes.

Yes resistance of tim2 does have an effect on heat transfer as you stated. Another reason I hate ihs's. They confuse the issue too much...

[aicjofs]

I don't have a lot of time to reply but

Maxxx I just used Paint Shop Pro because I thought it would be easier then making up numbers to satisfy what I wanted to say. Honestly I usually just use Excel which is rather limited.

As for what FLB is saying, I'd like to read Robotech's testing when I get a chance. Is it in this forum if I search or another forum. Sounds interesting. If measuring the side of the IHS, what was done to ensure that any warmer ambient air, if any, from power mosfets etc wasn't sqewing the reading?

I wish I had more time but I'll state a concern I have. How are we referencing TIM1? Seems as a side study people are concerned about the degrading nature of TIM1 joint. In fact a Intel CPU with the solder TIM would reduce variation a bit for this test. Surely Maxxx reseating the blocks 50 times is going to pump some "paste" out of that interface. Normally I'd say a few mountings would have a negligible effect(at least for the resolution this test is capable of measuring, or as much as CPU to CPU variation) as TIM2 junction would be far less repeatable and have a larger impact, but for the number of times we are reseating there needs to be some sort of baseline before and after testing to ensure it has a negligible effect on the results. Would a CPU at lowest multi, lowest HTT, and .9v be too much to fry a CPU without a heatsink/block? Have to consult white papers I guess. If you could reach steady state without frying the CPU at a known ambinet then you could see if TIM1 degraded over the numerous mountings and effected some of the blocks data.

I also wonder if the resolution ability of this test will allow us to say block 1 is better then block 2 which is better then block 7, or if due to accuracy concerns we will only be able to say blocks 1,4, and 5 are better then 2,3,and 8, which are better then 6,7,9 and 10.

[MaxxxRacer]

The next question is why is so much heat being spread through a relativly thin IHS. "My theory" is that it has to do with the high resistance (compared to the copper spreader) of the tim joint. As thin as the spreader is it's much easier for the heat to move lateraly than to move straight through the TIM joint. The more the heat spreads, the lower the TIM resistance becomes (due to the now larger heat transfer area).

The problem with your theory is that not so much of the heat is being spread through the IHS as you think...

[FLB]

As for what FLB is saying, I'd like to read Robotech's testing when I get a chance. Is it in this forum if I search or another forum. Sounds interesting. If measuring the side of the IHS, what was done to ensure that any warmer ambient air, if any, from power mosfets etc wasn't sqewing the reading?

Here's the link to the review

http://www.systemcooling.com/swiftech_apogee-01.html

robotech is a fairly smart guy so I assume he took proper precautions but it still would be nice to see a measurement from another source. Hopefully Maxx will be able to get a reading in the near future plus he will have a center IHS temp to compare it to which robotech didn't have.