Real Temp - New temp program for Intel Core processors

[unclewebb]

DJSUB: Based on E8400 testing and Intel using TjMax=100C for that processor, a Q6600 - G0 measures exactly the same so we'll have to assume TjMax = 100C for that one as well. Using TjMax=100C results in the core temperature being reported about 5C higher than what an IR thermometer measurement of the IHS shows. That is the difference that Intel wants us to accept though it might be a degree or two on the high side for the average processor. The Q6600 - B3 stepping as well as all of the original B2 series and the L2 series will be adjusted to TjMax=90C in the next release of RealTemp.

Freaky Freezer: I tried to hint that your problem is not likely a thermal paste application issue. Read the RealTemp documentation and try doing the calibration procedure and you'll have a better idea of how much each sensor is off at idle. You can have one sensor high and one low or both of them high or both of them low at idle by different amounts. The only way to figure this out is to do some testing. The bad news is that Everest and CoreTemp don't give you any way to correct for this problem. Since you like Everest, you have to get used to your idle temps being reported differently.

[Freaky Freezer]

The bad news is that Everest and CoreTemp don't give you any way to correct for this problem. Since you like Everest, you have to get used to your idle temps being reported differently.

I kind of figured as much, which is why I didn't bother with the calibration uncle. Maybe Everest in a new version will allow for a more specific adjustment of Tj. Max and then I can do the calibration.

Although since the re-application of thermal paste and setting Tj. Max to 100C, I'm reporting temps between 35c-37C and 35C-38C. A little better than the 4-5 difference before hand.

[unclewebb]

Here's some testing I did using the Intel approved TjMax=100C for my E8400. I also did this same test with a Q6600 - G0 with TjMax set to 100C and got the same results.

I played around with it for half an hour and even took it a little beyond TjMax just to make sure that it was thoroughly warmed up.



I tried testing at different temperatures and moved the IR gun around and around but the hottest it would ever show was 5C less than the reported temperature of the hottest core. This is the gradient that Intel says exists between IHS and the core even when the CPU is 99% idle. The above pic was taken 2 seconds after this one.



Once I put it back together I tried doing the RealTemp calibration test. The room temperature near the open computer was 21C and it was running at 1600 MHz and 1.08 volts. With these settings, at idle, an expected core temperature should be about 5C above the room temperature which would be 26C. This is what I get:



When using the correct TjMax, and no additional calibration, my core temperatures are being reported about 8C too high. This is a perfect example of "slope error." JohnZS has one core that reads too high and one core that reads too low so a combined error, when uncorrected, of 15C on his Quad is hardly surprising.

Intel finally releasing the correct TjMax for 45nm hasn't given the enthusiast community as much information as they thought they were going to get at IDF.

Here's a quote from the press release that was posted at Anandtech:

"Armed with this information, seasoned application developers and amateur coders alike will finally have everything they need to implement the most accurate, real-time core temperature display tool possible."

Well, not quite.

[Freaky Freezer]

forgive me uncle but you lost me. You said core temperature is usually 5C higher than room temp, which is 26C. So core temp should be 31C right? And you're showing 34C. So why is it 8C too high?

Also, my Smart Guardian is showing an Idle temp of 21-22 for the CPU, is it reading off the Tcase sensor? And if so, does that give any inkling as to what my core temps actually are?

[unclewebb]

Room temperature is only 21C so 21 + 5 = 26C which is 8C less then the 34C that is being reported.

I finally had a chance to compare my TCase reported temperature to the IR gun and it was 7C higher than what RealTemp was reporting with TjMax=100C. I've never trusted TCase readings from my board. Some boards are better than others but there's no way to tell by looking at it how well calibrated the TCase readings are.

[CompuTronix]

Unc,

Click on the link in my signature to use Section 9 (Calibrations), Part 1, in my Temp Guide to tweak an offset into SpeedFan, to correct for Tcase inaccuracy from BIOS.

EDIT: If you're running your Q6600 G0 (idle 16 Watts) and your Tuniq Tower (high-end cooler), then Tcase idle should be ambient + 3c. Check out Sections 6 and 14 to verify.

Also, note whether Tcase idle matches heat pipe IR measurements, where the pipes intersect the heat sinc. If you can find a clear line of sight, I think they'll be reasonably close. Since our Fluke's are +/- 1c, I'm interested to see if your Tcase idle observations agree with my findings.

Comp

[jason4207]

I kind of figured as much, which is why I didn't bother with the calibration uncle. Maybe Everest in a new version will allow for a more specific adjustment of Tj. Max and then I can do the calibration.

Although since the re-application of thermal paste and setting Tj. Max to 100C, I'm reporting temps between 35c-37C and 35C-38C. A little better than the 4-5 difference before hand.

Tjmax calibration and idle temp calibration are 2 completely different adjustments. Even if Everest allowed finer control of Tjmax adjustments you would still be left w/o any idle temp calibration (unless they decide to implement something in the future).

[JohnZS]

"Armed with this information, seasoned application developers and amateur coders alike will finally have everything they need to implement the most accurate, real-time core temperature display tool possible."

Well, not quite.

To me the whole thing was just a clever PR exercise for the upcoming Core i7 and all of us (you, me, Anandtech, all of us here on XtremeSystems etc) got sucked into this hoping to see details on the entire Core2 range's TjMAX (by stepping) and perhaps some hints and tips and how to accommodate for error.

At the end of the day all that Intel said in blunt layman terns was,,

"Yes we know our sensors in our 45nm Core2's are a little bit rubbish, here are a few TjMAX values for a couple of 45nm CPU's....if you are thinking of measuring thermal values there is a quantifiable amount of error in doing this and this error gets greater at idle temperatures.... BUT our up and coming brand spanking new shiny Core i7 processors come with working sensors, far more accurate and reliable than our current "Extreme" Sensors

Unclewebb do you plan on making 2.75 final anytime soon?
So far I am really impressed with it as my sensors all agree throughout the range (from idle to being at 100% load during stress testing).

Have you tried contacting intel asking if they have any further information to offer to assist you in developing RealTemp?
It would be nice if they gave you or at least made a public resource of all Core2 processors TjMAX values (by stepping to).

Keep up the good work, Kind Regards

John

[loonym]

Have you tried contacting intel asking if they have any further information to offer to assist you in developing RealTemp?
It would be nice if they gave you or at least made a public resource of all Core2 processors TjMAX values (by stepping to).

loloL... like intel really cares about devolping realtemp? They should be giving him loaner cpus to test.

[unclewebb]

I sort of gave up long ago on Intel being helpful to this project. I must have been dumb to think that they would say anything good about RealTemp which shows users that their temperature sensors sometimes "bottom out", aka. "get stuck". I wonder how many RMA's that has resulted in. Once Core i7 is out maybe one of their lawyers will contact me and let me know!

JohnZS: Intel's graph about the "slope error" was a little hard to interpret so I needed to do some more testing. I wasn't sure about whether core2 on my Q6600 and many other Quads suffered from TjMax error or slope error. Maybe a little of both.

I had to get that core over 90C before it started reporting the exact same as core3. That would indicate slope error and that error seems to cover the entire temperature range. If it was a difference in TjMax then that error would continue from idle up to and including TjMax.

The calibration method introduced in 2.75 is based on what my testing showed so I don't need to make any changes to that. I plan to bump up the original B2 and similar processors to TjMax=90C in the next release which should be out in a couple of days. I'll be calling it version 3.00 so users will know that there have been some major changes to TjMax and calibration and might even read the first line or two of the new documentation, maybe.

[emoners]

is the latest version 2.75?

[unclewebb]

is the latest version 2.75?

Yes. I've been doing more testing than programming lately.

[anzial]

Yes. I've been doing more testing than programming lately.

Where can I download 2.75?

[emoners]

Where can I download 2.75?

RealTemp 2.75 from this post few pages back

@unlce: thanks for the reply & great app

[anzial]

RealTemp 2.75 from this post few pages back

@unlce: thanks for the reply & great app

Thx!

[jarthel]

does anyone know the tjmax for e6xxx and e2xxx 65nm processors? did a search on the thread and nothing useful came up.

thank you

[unclewebb]

Intel still has not disclosed TjMax for 65nm desktop processors so all I can do is continue to guess.

Intel has stated that TjMax is 100C for an E8400. When I use that value, at higher temperatures where the amount of sensor error is minimal, the reported core temperature is approximately 5C greater than what an IR thermometer shows when pointed at the hottest spot on the IHS.

Using this new fact of life and working backwards, I have decided in the near future to start using TjMax=90C for the original B2, B3, the L2 and the E2xxx M0 series and probably a few others once I dig through the Intel specs some more. rge seems to agree that this is reasonable thinking based on Intel's published Thermal Specification for these processors.

Every program continues to assume that TjMax=85C for the original B2 but Intel has said that the logic behind that assumption is flawed because bit 30 in MSR 0xEE which is valid in the mobile processors to signal an 85C TjMax processor, is not relevant in the desktop processors. I'm getting ready to walk another plank and create some more controversy by introducing this never before used value in the near future.

I believe that the later E6x50 series are TjMax=100C like CoreTemp, Everest, etc. have been assuming for a long time.

RealTemp's original TjMax guesses were mostly based on IR thermometer testing. The new information released at IDF says that those numbers are too conservative and I should have added 5C to most processors.

Personally, I think that is a few degrees on the high side for the average processor but I've decided to more or less quietly comply with the small fragments of information provided to the user community from the guys in the blue suits.

[HDCHOPPER]

you had me all the way up to the word ... comply

[unclewebb]

Here's a nice write up from Gavin Steacy at Tom's Hardware about the IDF presentation and how these digital thermal sensors operate on the new 45nm processors. He even quoted the developer of RealTemp a couple of times.

I don't think I had my coffee yet that morning and was still a little upset by the lack of information actually released compared to the AnandTech pre-IDF story about full disclosure.

[Khalmyr]

just to clarify then...

the TjMax for the Q6600 G0 is 100ºC instead of 95ºC then?

[Ol'Baditude]

Here's a nice write up from Gavin Steacy at Tom's Hardware about the IDF presentation and how these digital thermal sensors operate on the new 45nm processors. He even quoted the developer of RealTemp a couple of times.

I don't think I had my coffee yet that morning and was still a little upset by the lack of information actually released compared to the AnandTech pre-IDF story about full disclosure.

Yup! I just read that.

From THW:
According to the developer of RealTemp, “[Intel] did not release enough information for any software developer to write an accurate program so we’re right back to guessing and making assumptions. It’s easy to take a pen and circle some numbers but they didn’t test, prove or show anything.”

So angry webb. lol.

just to clarify then...

the TjMax for the Q6600 G0 is 100ºC instead of 95ºC then?

I'm back to using 100deg. My temps are not what they used to be, but tbh, idling at mid thirties instead of mid twenties when the room temperature is 22degC. is quite a bit more credible. That, and my distance to TJMax numbers are now 5 deg higher, so who's complaining? Not me.

[JohnLoph]

Intel still has not disclosed TjMax for 65nm desktop processors so all I can do is continue to guess.

Intel has stated that TjMax is 100C for an E8400. When I use that value, at higher temperatures where the amount of sensor error is minimal, the reported core temperature is approximately 5C greater than what an IR thermometer shows when pointed at the hottest spot on the IHS.

Using this new fact of life and working backwards, I have decided in the near future to start using TjMax=90C for the original B2, B3, the L2 and the E2xxx M0 series and probably a few others once I dig through the Intel specs some more. rge seems to agree that this is reasonable thinking based on Intel's published Thermal Specification for these processors.

Every program continues to assume that TjMax=85C for the original B2 but Intel has said that the logic behind that assumption is flawed because bit 30 in MSR 0xEE which is valid in the mobile processors to signal an 85C TjMax processor, is not relevant in the desktop processors. I'm getting ready to walk another plank and create some more controversy by introducing this never before used value in the near future.

I believe that the later E6x50 series are TjMax=100C like CoreTemp, Everest, etc. have been assuming for a long time.

RealTemp's original TjMax guesses were mostly based on IR thermometer testing. The new information released at IDF says that those numbers are too conservative and I should have added 5C to most processors.

Personally, I think that is a few degrees on the high side for the average processor but I've decided to more or less quietly comply with the small fragments of information provided to the user community from the guys in the blue suits.

So that mean for E8400 the TjMAX should be set to 100 right ?

[unclewebb]

So that mean for E8400 the TjMAX should be set to 100 right ?

That's the official word from Intel.

[JohnLoph]

Thank you unclewebb (^____^)

[unclewebb]

burebista over on the TechPowerUp forums suggested adding an option to display the unmodified, digital thermal sensor data in the System Tray.

I'll upload a new beta version tomorrow with this feature.