Measuring AMD A64 Case Temperature

[DNA]

Has anyone used the technique outlined in the AMD Thermal Design Guide, Page 37, to monitor CPU temperature? If so, should the temperature measurements taken using this technique be considered more accurate than the CPU internal diode readings? What impact might the thermocouple hole/channel cut into a TEC coldplate have on the TEC's cooling performance?

link: http://www.amd.com/us-en/assets/cont...26633_5649.pdf

[JSimmons]

Do you have a link?

[DNA]

Originally Posted by JSimmons

Do you have a link?

http://www.amd.com/us-en/assets/cont...26633_5649.pdf

page 37, thermocouple instrumentation

[bigben2k]

I haven't seen anyone do that, yet. I've seen a few people use the old method, which involved cross-drilling a 1 mm hole in the baseplate of a HSF.

There's going to be an impact, for sure. It'll probably be small, but it'll be there. Assume a 1 mm thick IHS (there's very little data available on it). The round hole area is 5mm in diameter, and 0.75 to 1.0 mm deep.

It's interesting that AMD would recomend that; I would have hesitated to leave a 5mm round hole right smack in the middle of the core.

As for a TEC, I'd assume that you have a 1/4" thick coldplate? The impact would be smaller there. I'd bet on this method being more accurate than the mobo, but I'd try to keep the hole diameter smaller, or even better, don't make a hole, just use the 1mm wide channel.

The problem with mobo readings is that there is no calibration, and there's a wide variance in actual temperatures. You can assume a 10 deg C (up or down) variation. It works well enough to protect the CPU, for which it what it's intended.

[DNA]

Originally Posted by bigben2k

I haven't seen anyone do that, yet. I've seen a few people use the old method, which involved cross-drilling a 1 mm hole in the baseplate of a HSF.

There's going to be an impact, for sure. It'll probably be small, but it'll be there. Assume a 1 mm thick IHS (there's very little data available on it). The round hole area is 5mm in diameter, and 0.75 to 1.0 mm deep.

It's interesting that AMD would recomend that; I would have hesitated to leave a 5mm round hole right smack in the middle of the core.

As for a TEC, I'd assume that you have a 1/4" thick coldplate? The impact would be smaller there. I'd bet on this method being more accurate than the mobo, but I'd try to keep the hole diameter smaller, or even better, don't make a hole, just use the 1mm wide channel.

The problem with mobo readings is that there is no calibration, and there's a wide variance in actual temperatures. You can assume a 10 deg C (up or down) variation. It works well enough to protect the CPU, for which it what it's intended.

Thanks for the information. The TEC coldplate that I am using is ~0.35-in thick. Currently I have a 0.20-in metal digital thermometer probe embedded on the center point of the block thickness. The hole depth is such that the probe tip is located over the CPU center area. The probe/gauge was installed for the purpose of monitoring coldplate temperature which appears to work just fine.

My dilemma is that the motherboard/BIOS will not report CPU temperatures < 0C. To make matters worse, the motherboard/BIOS is suspected of reporting CPU temperatures 10 to 20C higher than actual, per Abit AN8 Forum user comments.

Can one extrapolate a reasonably accurate CPU temperature from the coldplate temperature that I can measure, +/- 1C?

If not, would you recommend the AMD thermocouple technique using a milled channel only as being my best option aside from a calibrated internal diode reader?

[bigben2k]

There's not going to be an offset from what you measure; the difference will depend on the amount of power that the CPU puts out.

I don't think I'd be able to calculate any offset either; the calculation is going to depend on a lot of variables.

I'd take a guess at (load, 75 W) 15 deg C.

Check out this thread for some info:
http://forums.procooling.com/vbb/showthread.php?t=9445