Laser temp probe directly to CPU and coretemp (pics)

[rge]

ambient-----Probe temp CPU--------Coretemp 95 reading
24C-----------28C (no pic)---------------18C (no pic)
24C-----------31.6C (pic)----------------27C (pic)
24C-----------43C (pic)------------------53C (pic)
24C-----------58C (no pic)---------------75C (no pic)
computer shutdown last coretemp read 80C, but mouse and desktop froze first, so who knows what temp was.

I positioned the monitor with coretemp at top left, so I could get a pic of both coretemp 95 and my laserpro directly hitting the back of the CPU (note water block is removed and hanging down).

I held the block on with fingers, booted, initial and final measurement I could not capture, camera pic delay too slow and temp rise too fast. The other two I was able to get pics of.

Clearly DTS (coretemp reads from) is either non-linear, or simply does not read accurate at low temps. I think it is possible that coretemp could be accurate at temps around 50C+, but no proof it is. I was adding 15C to temps thinking that was accurate, now I dont think that anymore.

[newls1]

Hmm, interesting

[boogle]

You've got an E6850 there, doesn't that have a Tjunction of 85c? Which should mean the Tjunction is being picked up correctly - and the CoreTemp is 'accurate' without the need to add 15c?

Either way, these results are very, very interesting.

[rge]

Core temp is set at Tjunction of 85C for E6850, I dont think anyone knows for sure if its correct, but it might be.

One thing I learned in playing with this for hours, before taking pics, heat is conducted very fast. With waterblock off and reduced clock speed and vcore undervolted, coretemp changed from 19C and reached equilibrium at ~56C in about 4 seconds, temp of back of CPU scaled over same 4 seconds and reached equilibrium at ~45C. Both coretemp and back of CPU had reached equlibrium in same 4 seconds, and scaled at same speed. Put heatsink on, and less than two seconds later temp back to 19C on coretemp, and back CPU measured 28C. Heat gradient appears to occur instaneously.

[AndyM]

Wow, nice research! Good to see testing like this being done.

[AndyM]

Could you do a comparison with TAT and laser thermometer? I noticed that coretemp reads low at low temps and high on high temps in comparison to TAT. If your thermometer and TAT tracked closer, that would be good to determine the preferred temperature app.

[Bail_w]

in my case, my Xeon 3210 (tjunction 100 degree c) full loaded at 60s with 1.26v when i touch the radiator i did not feel any heat at all. on the other hand, my E6400 (tjunction 85 degree c) full loaded at 65s with 1.57v when i touch the radiator it feel warm. I would like to see some results with a quad.

[lawrywild]

What about the temps your motherboard reports the cpu as?

[lad]

The temp diodes inside each core are at the hottest part of the cpu's. Your IR thermo was measuring the temp of a large area of the heatspreader.

Unless you allowed the temps to stabilize, and then accounted for temp differential throughout the chip, the IR thermo won't tell you anything useful.

And the *good* ($200) IR thermos aren't very accurate - 5% or 5c, whichever is *greater*.

[Bail_w]

What about the temps your motherboard reports the cpu as?

temp of the motherboard reports always about 10 degree lower than coretemp.

[rge]

I played with TAT, speedfan, everest sensor, coretemp and cpuz on most of the time. I believe temp2 on speedfan is the mobo cpu temp, but reports same as everest cpu mobo. I have my rig back together now, with reseated cpu, and tired of testing...but...some scribblings

TAT read 0-2 degrees higher than coretemp at lower temps, both were reading below ambient. At higher temps TAT read 1-4 degrees lower than coretemp.

One reading at higher temp was
ambient 24c
probe to cpu 49C
coretemp 95 core 61C
speedfan temp2/everest cpu 68C
TAT core 58C


undervolted to 1.15 volts vcore, 6 x 200 or 1.2Ghz cpu
wait 4 minutes to steady state.
probe to cpu 46C
coretemp 55C
did not have speedfan running or did not record.
TAT 53C

At core temps from 60 to 70..
Coretemp read ~14-18C higher than probe to CPU
TAT would read ~12-15C higher than probe to CPU
cpu temp read ~18-24C higher than probe to CPU

So who knows what is accurate, though the cpu mobo temp was sometimes all over the place, 82C when that was clearly wrong, and 35C (with coretemp 60C) and that was clearly wrong. These weird cpu temps occurred when settings were changed without cold reboot after bios.

[rge]

The temp diodes inside each core are at the hottest part of the cpu's. Your IR thermo was measuring the temp of a large area of the heatspreader.

Unless you allowed the temps to stabilize, and then accounted for temp differential throughout the chip, the IR thermo won't tell you anything useful.

And the *good* ($200) IR thermos aren't very accurate - 5% or 5c, whichever is *greater*.

Only thing I was trying to show, is that the DTS seems to be variably accurate at low vs high temps.

But yes, no way to measure the core, just the back of cpu.

My laserpro is supposed to be within 1C of accuracy. My wife makes cosmetics, and I used her very accurate and expensive touch temp probe to check mine, mine was within 1-2C of hers at several different measurements. In fact underneath my tongue just now reads 98.4F/36.9C, body temp is usually 98.6.

Regarding the temps to stabilize, I thought the exact same thing, till I started playing with it. You will have to do it yourself, before you believe. But if you read my second post, equilibrium is almost immediate. The scaling between laser pro to back of cpu and coretemp is immediate, and when core reaches steady temp, back cpu temp is steady, happens immediately.

[lad]

[QUOTE=rge;2364885]Only thing I was trying to show, is that the DTS seems to be variably accurate at low vs high temps.
...
QUOTE]

Interesting. Wish my IR gun was that accurate.

Thanks for the extended explanation - much appreciated.

[fornowagain]

Its been speculated for awhile that the DTS is fairly inaccurate at lower temperatures. It is used for TCC throttling so not entirely surprising.

Check this thread.

Nistomax: Here's a good example why I don't put too much faith in the CPU temperature reported by most motherboards.



13C is a great temperature but in a 22C environment with an air cooled processor it's impossible. SpeedFan reads this sensor and reports it as CPU but it's obviously meaningless.

For my E6400, when software guesses that TjMax=85C, I think the DTS is providing a pretty good approximation of what my absolute core temperature is from say 45C to 85C but I'm not convinced that the DTS is accurate at lower temperatures. It was designed and calibrated to be accurate at the throttling point but I have no data to prove how accurate it is at low temperatures like you're running.

Even if it's not accurate for reporting absolute temperatures, it is always accurate at reporting relative temperatures. If CoreTemp 0.95 reports that your Delta to Tjunction is 80C one day and then you do a few mods and it reports that your Delta to Tjunction is 85C the next day, your processor is definitely running 5C cooler than what it was running at before.

The DTS was designed to report how far you are away from TjMax and for that purpose, it's fully documented, calibrated and it's the best temperature information available for users.

[bing]

Doing it the correct and Intel's way DOES REALLY matter !


Ever wonder how are those Intel CPU engineers measure Intel's own CPU in their uber leet lab ?

Warning, cpu pron pics inside, not advisable for minor !

Intel® Core™2 Extreme Quad-Core Processor and Intel® Core™2 Quad Processor Thermal and Mechanical Design Guidelines Download the PDF file.

For those who likes quick shortcut, just go to page 81 and beyond !

After you've read that and if you believe it, next time you will start chuckling & gigling back there when you hear these kinda comments :

"TAT cpu temp reading rocks ! I ignore others"
"Coretemp is the best and I believe it !"
"Temp diode reading is the most accurate one"
"The Tjunction should be xxx, not yyy .... "
..bla..bla..

For those non believers even after you've read it, c'mon, if this Intel's document is not trustworthy, then what else ?
That program that shows the temp that you "want" to believe ?

Conclusion ? Software based CPU temp reading is NEVER ACCURATE, period !

[rge]

That article is interesting. They are measuring Tcase directly, instead of a guesstimate based on power dissipation?, which I think is the Tcase software measurement from reading (but could be wrong). But intel does use DTS (read via software) as accurate relative temp to throttling, just can not convert to absolute as you point out.

Other tech articles
http://download.intel.com/design/pro...x/31780401.pdf
ftp://download.intel.com/design/proc...s/31327805.pdf

In looking at figure 3-1 of the first of two listed above and article you referenced, by gunning back of cpu (IHS) I should be reading Tcase or within 1mm of copper of it. But intel's stategy allows to read with a cooling block and tim attached.

Wonder how hard it is to calibrate one of those probes and cut that slit in cpu. Though I would not try that until I upgrade in year or so.

[rge]

Does anyone know what CPU temp is obtained from in speedfan and everest. I thought from my previous post it was calculated from power dissipaton? but obviously not the case. Did some more playing around.

At 1.2Vcore 200x6mult or 1.2Ghz, cpu core temp increased finally to about 65C, speedfan cpu (?tcase) measured 70C. MY laser pro measured 47C. When I touched the back of cpu it burned my fingertip. Somethings not right. First if Tcase is calculated temp from power, it would not know my waterblock is off, so maybe it is measured??? anyone know. Secondly 47C should not burn my finger.

I know my laserpro was accurate on other surfaces, then I remembered Unclewebb had done this and found that laser did not accurately measure temp on cpu because shiny surface. So I put a piece of tape on measured again, and decreased vcore some more, so I could leave things on for awhile without getting to hot.

ambient 24C
idle and waterblock off
1.12V 1.2Ghz steady state after several minutes
speedfan/everest cpu (?tcase) = 65C
coretemp/TAT read cores = 58/56
laser pro no tape 47C (but burns finger)
laser pro with thin tape 64C
Finally couldnt stand it any longer, wifes asleep, used her accurate touch probe temp read 63C

Played around some more, definitely need tape for gun on cpu to read accurately.

But funny thing, after my poopooing Tcase temp, actually cpu temp by speedfan (?tcase) was within 1-2 degrees of her accurate temp probe or my laser gun with tape to back cpu at several temps. I will try to get pics when wife awake, but that will be with tape and my laserpro, not her temp probe, since she has to snap pics why I hold and I enjoy living.

Anyone know how cpu temp mobo is obtained? Is it supposedly to case? Calculated or measured? Still wondering how I sometimes get such bizarre measurements from it though, and this time seemed accurate for tcase.

[bing]

Hint : Try PECI ON and OFF both, observe the "CPU" and the "Cores" temp at SpeedFan on both states, of course do observe with the laser pro as well, and make brief summary !

[rge]

I tried to look for peci before, I dont think it exists on my gigabyte p35 DQ6, if anyone knows if it exists or how to find it on that mobo, I would love to run that test.

But since I now have accurate readings on back of my cpu, which should be close to Tcase actual (via putting thin tape on cpu and double checking multiple temps with very accurate temp probe).

I found something very interesting. Summary with pics attached.

laserT back cpu------Speedfan CPU------coretemp
------45C----------------42C---------------34C
------51.9C--------------50C---------------42C
------64.7C--------------65C---------------58C
------66.8C--------------66C---------------60C

So now is Speedfan accurately measuring Tcase at temps above 45C? I tried this for hours, tracks almost perfectly all the time, hundreds of measurements, providing temps are 45C and above.