Real Temp - New temp program for Intel Core processors

[CrazyNutz]

You could go to the first post and follow the directions religiously.

rge: TjMax is documented to be 100C for the laptops but now I wonder if they suffer from the same inaccuracies as the desktop DTS sensors do at low idle temperatures. They most likely do.

I think there is something that must be clear, from what I can tell Intel never
Intended the DTS sensors to provide us with accurate temps, we can obviously read the data from them but the data they provide is to let us know how close the proc is getting to thermal shutdown. So they are mainly for fan speed control, and alerting of catastrophic cooling failure. But It
is still nice to get a semi-accurate core temp to give us a decent idea how hot our procs are getting when we are
trying to squeeze every little oz of Mhz from them

[Ace-a-Rue]

Please, anybody, host the "RealTemp.zip" somewhere else. That link does not work .

i'll post it up on my website...STANDBY

[Ace-a-Rue]

Please, anybody, host the "RealTemp.zip" somewhere else. That link does not work .

i'll post it up on my website...STANDBY

here is an alternate place to grab unclewebb's file: https://home.comcast.net/~captain-is...d/realtemp.zip

[jason4207]

If Tjmax on my E8400 is 95, then there is no gradient from core to tcase on full orthos load, as my calibrated cpu temp is now same temp as core. And I can show for example, that when speedfan cpu shows temp of 53, IR to back of cpu will be same.

Unclewebb, you could easily calibrate your cpu, if have not already, and test same, would be interested in what you get. When I reach 105C core, my tcase measured by IR was 94.8, so I got the same result as you. The only issue is, does any gradient exist between core and tcase at that time. Your control chip, one did not. On my E6850, a gradient of 5-7C existed at all times, and on my E8400 a gradient of 10C existed at all times.

rge: I did all my IR testing with the CPU at idle. Even when it was over 100C it was still only at idle. Temperature gradients between TCase and core temp exist but I believe that when at idle, the difference between the true core temperature reading and what you can measure on the top side of the IHS is so small that it is within the accuracy of the test equipment espicially at lower temperatures where I was testing.

That's why I dropped TjMax for the E6850 from 100C to 95C and dropped the E8400 from 105C to 95C.

My initial thinking is that the temp gradient b/n the die and the IHS will be higher at higher temps relative to ambient regardless of idle or load state. You said there would be a small gradient at low temps. I agree, but I don't think 100*C is low whether you're idle or load.

At what temp reading on the IHS did throttling occur? If it was at 95*C then I'd think the die would be a bit hotter, say 100*C, and throttling occurs ~5*C from Tjmax (I think), so that would make Tjmax 105*C.

If throttling occurred at 85*C Tcase then I can see Tjmax being 95*C.

Absolutely not. A silicon die is not a very good thermal conductor (bulk silicon ~148 W/m.K and far worse for others materials used to create transistors structures, etc.). Thermal gradient within the die can achieve 20~40+ °C between the hottest and the coldest point, depending of the load. That's why a die presents what we call 'hot spots' because there are hotter than the rest of the die (more power density such ALU or fetch units for instance) and they are the major problem to cool the chip and ensure reliability. Die temperature map is and won't be uniform at all, ever !

On a dualcore, it's clearly seen. Load only one core with TAT for instance and look values reported by the DTS sensors, there will be ~4-7 °C between them (depending of cooling system) and however the sensors are only ~5 mm far from each other. According to you, the sensors should be at the same temperature, which is nonphysical. There's always a thermal gradient when heat has to travel through materials. The worse the material is, the higher the gradient will be at a given distance, Fourier law. CPU TEMP provided by the third thermal diode don't have to be the same than DTS TEMP, it's located between the 2 DTS sensors but in a 'colder' place (results are affected by mobo circuitry unfortunately, but a external system could normally be used to calibrate and use it like on AMD CPU).

Nothing is linear here, you can't guess the difference between sensors, that's why temperature difference between them is evolving for each situation (idle/load but what load...). Intel already showed the large deviation between Tj provided by DTS and Tcpu provided by the third sensor depending of die loading map (power map more precisely), there's no relation at all between them ! We can see Tcpu like a 'mean' temperature if we want, that's all.

There's always a gradient when dissipated heat travels from one point to another, it's physical. And please stop using IR guns, there are the worse instruments often used by people like a toy. It requires some cautions (like IR cams) because of the emittance problem of the surface we want to scan (why the yellow adhesive tape near the IHS on the first page? What's for? TC bonded to IHS?). Without preparation, pointing an IR gun on a +/- shiny surface like IHS or die is not good, the error will be higher because of radiation dispersion (gun don't measure temp but radiation) and gun will tell you a lower temperature than reality. If not adjustable, the gun is generally tuned for a 0.95 emittance, it's useful for black/matte surfaces (near black body) but not reflective/shiny/clear ones. Using a thermocouple bead is more useful and confident. You can also paint the IHS with black paint to get a emittance near maximum during measurements, that is 1. Moreover, you can't conclude anything with measuring Tcase (at the IHS center) for Tj or Tcpu because of the thermal gradients (a silicon die is ~0.04 °C/W through its thickness and there's TIM1 and IHS in the heat travel to take in account too for desktop CPU).

What matters is the real Tjmax, nothing else. Unfortunately Intel keeps its secrets like many others and it's a bit strange to not reveal such 'non critical' information I think... It's near 100 °C each time, that all. A bad guess and you have your cores under ambiant temp, so you know there's a flaw in that value. The absolute value will be false, but not the relative one. That don't make any great difference if the CPU is stable and not throttling, even with an o/c. What is observed here is related to the fact that nothing is linear again. You are pointing an IHS when its surface could have a 10 °C temperature difference under load, IHS temp is not uniform at all. That's a lot of flaws and errors at the final in the process...

You bring up a lot of good points.

The gradients that Intel shows in their diagrams is with a processor under load. All my testing was with 0% load so it's my belief that whatever gradient there is, is minimal and within the accuracy of the equipment being used.

With the core shrink of the new 45nm chips the gradient from core to core is getting smaller at full load because the on chip DTS sensors are getting closer and closer together.

The reason I use masking tape on the core is of course to reduce shine and to make the readings repeatable. With my computer unplugged and sitting over night I find that when I do an IR reading on the taped core, it is exactly equal to the ambient temperature. To me that says that temp readings through a piece of masking tape are legitimate.

I've had these same arguments with others and I realize I'm not likely to convince you. My question for you is how do you think Intel calibrates these chips on the assembly line. They have better testing equipment than I do but their methodology has to be quite similar. Get a reading from an IR thermometer and create a calibration curve to equalize measured core temp vs what the DTS is putting out.

Here's a good example of my E8400. I'm running Prime95 small FFTs, first on core0 with nothing on core1 and then I went into Task Manager and swapped it over to core1 with nothing on core0.

When the full load was on core0, it was averaging 47C while core0 heated up to 46C even though it wasn't doing anything. With the load swapped, both cores were now running pretty much the exact same temperature. The overall average difference is now less than 1C with 45nm.

I don't believe for one second that an IHS reading with both cores at idle is going to be 10C or even 5C away from what the actual core temperature is.

I have to disagree until more evidence is presented to the contrary. For all we know the DTS sensors are very close to each other on the die. The IHS has to be cooler, though.

That is interesting, especially how well they track. I was going to argue that maybe you were in the temp range where accuracy was less reliable, but instead I tried with linpack (finally a use for it) myself with rad fans off, got same results at higher temps, setting affinity to core 0, then core 1.

Regarding accuracy of IR, I wish 6 months ago, when i used my wifes accurate temp probe from work to double check IR through masking tape versus temp probe to cpu, wish I had taken a pic...would just flash that up each time as they were within ~1C at several different temps. Though, I used the probe thinking it was not going to be accurate.

You almost made a believer of me, using your useful program regardless, but would be nice to try it on a mobile cpu with known tjmax.

I'd like to see that as well.


Overall I think you've done a fantastic job w/ this program, but I still think it needs to be tweaked more. I think there should be a curve between the lower temps and higher temps instead of 2 linear lines that intersect. And, of course, I think there is a temperature gradient b/n the die and the IHS that has to be taken into account...especially at the higher temps (100*C), and even if at idle.

[alcatraroz]

hello unclewebb...I TRY MANY TIMES the sensors calibrate but the results change continously...and when i raise the vcore the movements get up but everytime show differences between cores 0 is ok but 1 always lower.. its correct?

[zillaoc]

Here's another download linky, This is real temp and the ia32 driver in one zip.

Download Link > BiTTDaily.com

[Darth Vader]

Ace-a-Rue
Thanx

[rge]

Can anyone confirm what cpu temp is from speedfan/everest etc., given any modern mobo and E8xxx, ie E8400. The E8400 has two DTS sensors, one for each core, and no diode. I am assuming software reporting cpu temp is reading signals from a mobo diode sensor directly underneath and in contact with the metal of the cpu socket, so it is approximating "Tcase" but reading through the underside, thus through ?slightly different makeup of materials?

[TheKarmakazi]

awesome program. great discussion too. although im not technically knowledgable enough to contribute to the debate, i am certainly learning alot. thanks fellas, keep up the good work!

another thanks to unclewebb also for creating and sharing such a great tool!

[WiTcHkInG]

Great work unclewbb.

followed the link Felix posted and there is a x64 driver there. Somebody else has used it to make temp program (here).
Maybe you could use it to get yours going on X64 also

[Viss]

Good job uncleweb seems to work like a charm, thanks a lot

[unclewebb]

Here's a quick update.

Version 1.1

-fixed the Xeon 4300 bug as well as greatly improved processor recognition. If yours doesn't work, let me know and send me a CPUz screen shot as well.

-disabled early Pentium 4 support since it didn't work but left 65nm Pentium 4 support enabled. Someone will have to let me know if this works.

-added quicker temp update after a quick bench as well as a checksum feature.

-deleted my judgment of your DTS sensors (OK-Bad). Better documentation after a SensorTest so you can judge for yourself what is good or bad.

-trying the File Den now for hosting. Seems fast.

http://www.fileden.com/files/2008/3/...7/RealTemp.zip


Things to do:

-add system tray support and maybe even x64 support!

[zillaoc]

Here's a quick update.

Version 1.1

Just updated my link to it too.

Download Link > BiTTDaily.com

.
.
.
.
.

[Ace-a-Rue]

Here's a quick update.

Things to do:

-add system tray support and maybe even x64 support!

COOL!

[unclewebb]

I think there is something that must be clear, from what I can tell Intel never Intended the DTS sensors to provide us with accurate temps,....

From my testing I would say that is only half true. In the upper range within 35C of TjMax, the DTS seems to be very accurate and moves step for step with core temperatures changes.

It's only at the low end where it isn't 100% accurate for reporting idle temperatures. A simple correction factor can get it very close to reality, even down low.

[Kilyin]

Regarding x64, this may or may not interest you, but in Vista x64 to get CoreTemp to work I had to use a file called atsiv.exe in a batch file to load the ALSsysIO64.sys driver and launch CoreTemp.

[unclewebb]

Thanks for the info.

Can anyone confirm if this program works in Vista or XP x64,
http://shefot.com/toni/fajlovi/c2dte...inst_e_x64.msi

It uses the CrystalCPUID driver and might be the easiest way to share RealTemp with the rest of the world.

[leffe]

Thanks for the info.

Can anyone confirm if this program works in Vista or XP x64,
http://shefot.com/toni/fajlovi/c2dte...inst_e_x64.msi

It uses the CrystalCPUID driver and might be the easiest way to share RealTemp with the rest of the world.

Confirmed. Although it shows idle temps around 60 when I'm cooling with water and U-guru shows ~30. Using Vista64.

[unclewebb]

jason4207: I have to disagree until more evidence is presented to the contrary. For all we know the DTS sensors are very close to each other on the die. The IHS has to be cooler, though.

Of course the DTS sensors are very close to each other. The core is about as big as your thumb nail so the large gradients that used to exist in the Pentium 4 era have been reduced to next to nothing. If one core at full bore brings the other core up to within 1C of it then at idle the difference between an IR gun IHS reading and the "real" core temperature can't be greater than that.

You don't have to go all the way up to TjMax when testing. At an IR reading of 60C, the DTS shows 35 on my E8400. at 70C is was showing 25 and at 80C it was showing 15. I plan to do some more high temperature testing when I get a chance to make sure the PROCHOT# feature is working properly.

Confirmed. Although it shows idle temps around 60 when I'm cooling with water and U-guru shows ~30. Using Vista64.

Good news. That program is displaying the actual DTS reading and is accurate.

[Spacehead]

Of course the DTS sensors are very close to each other. The core is about as big as your thumb nail so the large gradients that used to exist in the Pentium 4 era have been reduced to next to nothing. If one core at full bore brings the other core up to within 1C of it then at idle the difference between an IR gun IHS reading and the "real" core temperature can't be greater than that.

You don't have to go all the way up to TjMax when testing. At an IR reading of 60C, the DTS shows 35 on my E8400. at 70C is was showing 25 and at 80C it was showing 15. I plan to do some more high temperature testing when I get a chance to make sure the PROCHOT# feature is working properly.



Good news. That program is displaying the actual DTS reading and is accurate.

Speedfan 64/64
That program -21/-21 (Tjuntion 85)

Speedfan shows correct temp.

[ixtapalapaquetl]

Does anyone know anything about the "Biot number"? I was describing the sensor issue with a friend schooled in thermodynamics and I made the following analogy:

Imagine a bathtub filled with water and multiple identical stones that have been heated to the exact same temperature. Now imagine placing a thermal sensor at the far end of the tub and another above the drain. If one hot stone is placed on top of the drain, the reading on the near sensor will change rapidly, but soon the other sensor will report a change as well. If two stones are placed in the tub, both sensors will read higher than when one stone was placed in the tub. More stones, higher readings. My claim is that this will happen very predictably - as long as we are not (for example) moving the sensors' locations or stirring the water. We could seemingly take a reading from either sensor independently and determine exactly how many stones have been added for they are both measuring the same thing - namely the heat from the stones.

The sensor above the drain would obviously represent one of the core sensors, and the one on the far side of the tub the motherboard sensor. Both sensors measure the heat generated by the core, one near and one far.

His response was that this could be true given that the Biot number was less than 0.1. If this were the case, the system could effectively be treated as a single blob and the analogy would be valid. Further, it would also confirm the two core findings described by unclewebb and others. I am having difficulty, however, finding silicon's heat transfer coefficient. Any gods of thermodynamics care to comment?

[tonyptony]

Anybody tried it yet with an E21xx or E2200? M0, specifically?

[Ace-a-Rue]

Confirmed. Although it shows idle temps around 60 when I'm cooling with water and U-guru shows ~30. Using Vista64.

Good news. That program is displaying the actual DTS reading and is accurate.

hey, unclewebb...did i read the above para correctly?!...Uguru chip on Abit mobo's read from the DTS sensor?

[OM DUST]

If I am at full load P95 3 Hours at 57-60c and I have 30-35c left to Tjmax than is it safe to keep pumping some more volts into my CPU? I'm at 1.425 in BIOS already and was reluctant to push further.

[Grendel66]

Excellent work ! Too bad my cores are both stuck at 45C (E84000). I think I saw one move to 46C once..