Real Temp - New temp program for Intel Core processors

[rge]

Intel just deleted my post again, this time in a matter of hours. Just dont get why they would delete it. Refusing to answer, sure.

KTE, I accept any error from measuring "Tcase" at surface versus 380 microns deep as very small, whether right or wrong, primarily because no heatsink is in place. No question with a heatsink, it would be necessary to place the thermocouple below the surface. And actually we have no interest in measuring tcase anyways, so I guess we should say we are measuring the casing of the cpu, at idle, UC/UV steady state with no cooling solution on, as the casing/core are near equilibrium at that point, ie within ~0-2C.

But I completely agree the main issue boils down to the gradient between tcase and tjunction at idle, and the one I have the least info on is across TIM1. No one has proof but intel, however you can find some evidence in several different ways.

Mathematically, you can guestimate it using known values with other tested materials in intel papers, especially the one I quoted when I made my "assertive question" to intel. (Which they have now deleted again.)

There were two experiments that indirectly measured it (neither by intel) across TIM1 and IHS. One was talked about in a discussion on overclock.com, someone named bill and? and they stated the gradient was 2.5C across that of P4, and they stated "we all agree on that", these were individuals that design heatsinks for a living, but they offered no proof. But they vigorously argued about everything but the 2.5C gradient. And TIM1 from that time til now has increased in conductance ~10x, and a slide show by intel development team has stated decrease in resistance, have to find the slide presentation, (Edit..found slide presentation...slides 35-39, http://download.intel.com/technology...Packaging.pdf). The 10x is from looking up those that sell TIM1 for cpus and comparing their thermal conductance, to those mentioned in previous studies, not only from older P4, but even much higher than the slides mentioned above in 2005.

Also, it was kinda measured at idle, in paper below, though this compares measuring casing temps underneath, though intel has stated 2/3 heat travels via TIM1 to this point, and 1/3 directly to casing below...so not direct analogy, but they found only 0.4C gradient at idle. http://www.overclockers.com/tips443

Other evidence is just from reading a lot of papers, and them saying things like 20-30% gradient across tim1, because although thermal conductance is higher for tim1, tim1 is only ~75-100um thick, and lower thickness directly lowers thermal resistance....again just mathematical guestimate.

So my best guess, using all info, is that gradient is 0-2C, under low power, UC/UV idle, but my question was to intel because as you state, only they know for sure. But given all I have read, I have a hard time believing at those conditions it is much higher, especially when no heatsink is applied to drive the gradient.

[jason4207]

Shows the same exact temps as Core Temp, and Everest without any calibration both idle and load for the E6600 (TJ = 85C) I'm using. The minus/minus and plus/plus calibration will deviate 2-3C at most. So I guess I've got a pretty accurate sensor?

read through the post and if I have an e8400 I should just leave the calibration in the middle correct?

OK guys...how many times must we ask this question?

The (++) (--) Idle Calibration feature is mostly for the fanatics amongst us that go ape s-h-i-t when they see an idle temperature that is a couple of degrees below their ambient temperature. Some users see that as a sure sign that TjMax must be wrong but it usually isn't. If you run your E6600 at a very low core voltage of 1.10 volts and underclock it to 1600 MHz (266x6) then it will likely confirm that your idle temps are too low. A (+) or (++) calibration setting is often times best for the early Conroe Revision B2 processors.

And the answer also surprisingly shows up in the 1st post.

Idle Temperature Calibration:

If you are interested in this feature then you need a way to find out if your processor reads too high or too low at idle. The method I use is I set my processor to run at a front side bus speed of 266 MHz and then I set the multiplier to 6.0 which results in approximately 1600 MHz. I also lower the core voltage as low as it will go. My Asus board goes as low as 1.08 volts which is fine for this test. You could also use SpeedStep and set C1E to on in the bios to accomplish the same thing. The goal is to reduce the heat output of your processor to a minimum.

By reducing your processor down to a common fixed value, I've found that a well air cooled processor's cores will idle at approximately 3C to 5C above the ambient temperature. If your reported idle temperatures are going below the ambient temperature then you need to use a correction factor of (+) or (++) to get them up into a more believable range. If your CPU is like my E8400 and you notice reported temperatures that are significantly higher than the 3C to 5C range then you'll need to use a (-) or (--) Idle Temperature Calibration factor to bring them down to a reasonable range. That's all there is to it. As long as TjMax is chosen correctly and you do a simple calibration, your reported Core Temperatures will be very accurate and comparable to other users who are using the same software and have also individually calibrated their processor.

[Demo]

I just tired using this program. Can someone tell me why my temps are so different from everest 4.50.1330? isn't the tjmax suppose to be 105C for the qx9650?

Not 100% sure, but people believe its 95c based on some findings.

[unclewebb]

isn't the tjmax suppose to be 105C for the qx9650?

Can you post the Intel documentation for that? Just because a temperature monitoring program assumes what TjMax is for a processor doesn't mean that it is right, including RealTemp. Intel provides no public documentation for their desktop Core processors so all programs are left guessing.

I went with 95C for the QX9650 because that is what the dual core 45nm desktop processors are using and a Quad core is basically two dual cores joined together. I know 105C is correct for the mobile 45nm chips but I've yet to read of any proper testing on a desktop 45nm Quad core chip.

If you are convinced that TjMax=105C then you can still use RealTemp. Just go into the RealTemp.ini file and set TjMax=2 and that will bump up your TjMax two 5 degree notches from 95C to 105C. Problem solved.

KTE: I was basing my maximum temperature calculation on the last picture you posted.



RealTemp can read the on chip DTS data OK. If you compare core0 vs core1 you see that DTS=114 on core1 so RealTemp calculates the temp to be 114C. Core0 has a DTS=100 reading so that implies that core0 was 14C warmer or 128C. I realize that's a sudden jump up from what you were observing and since I wasn't there watching, it's impossible to say what happened. It's also possible that any interpretation of the DTS data is meaningless up here. It's certainly not worth converting a good Quad chip into a key chain.

I was planning to do some of this extreme kind of testing using my E2160 but it was a challenge getting it hot enough, even without a heatsink and fan. Running Prime without a HSF is not really Intel approved! At the moment we'll call it KTE 1 - RealTemp 0. In theory, the latest version of RealTemp should be able to keep up but I'll have to wait until I can do some further testing to prove it. I'm in the same boat as you, I need my computer up and running at the moment but I promise to report more of my real world testing in the future so stay tuned. I really appreciate all of the extreme testing you've done.

[flexy]

let me get this straight, i cant think right...

With "ambient" temperature you mean the real room ambient temperature?

So...my real room temp is 21.5 right now and realtemp reports me 19 19 23 23 on the cores.

Since 19 C on a core is physically impossible (due to my ambient temp) i need to calibrate higher...preferrably 3-5 deg higher what ambient says?

Am i the only one seeing a problem with the guessing and then "calibrating" according to the guessing ("should be 3-5 higher than ambient...")...

As of right now i can only assume that your statement ist right "3-5 higher than ambient"....doing idle calibration right now on G0 w/ 1.08Vcore and 1.5Ghz, lowe FSB.

G.

[unclewebb]

Am i the only one seeing a problem with the guessing and then "calibrating" according to the guessing ("should be 3-5 higher than ambient...")

It gets worse flexy. All of my testing has only been with the dual cores so far. It's likely that my 3C-5C above ambient guideline will need to be boosted up a couple of degrees to be relevant for Quad cores. I have a G0 Q6600 in stock awaiting a thorough work over so be patient. RealTemp is only a month old so with more time and more testing by myself and other users it can only get more accurate. Post what ever you learn about your Quad.

If people are upset by programmers having to create their own test methods and guess then they should write to Intel and tell them that hiding all details about the digital thermal sensors and TjMax for their processors is wrong. I don't understand what the big secret is all about.

[GripS]

Appreciate the work. I noticed a 2-3C discrepancy between this and coretemp.

[Kyo`]

RealTemp 2.1 is reporting almost the same temps in idle as CoreTemp for my Q6600, after the ++ calibration (1C difference on some cores). Load temps are 4C lower on each core. So working fine for me I guess, keep up the good job.

[loonym]

unclewebb, I really think my q6600 is either very close to correct or requires - for proper idle calibration. However it's not really idle temps that require monitoring
I am really interested in tjmax since that number being reported correctly seems to equal realtemp reporting load temps correctly. Is this so?

[unclewebb]

I am really interested in tjmax since that number being reported correctly seems to equal realtemp reporting load temps correctly. Is this so?

If TjMax is accurate for your processor then your load temps will be extremely accurate. My chart in post#1 shows what I've found. It's at temperatures farther away from TjMax where the DTS data coming out of Core processors does not seem to be changing at the same rate that the core temperature is changing at so temp errors begin to accumulate. A program that guesses wrong at TjMax or manipulates it higher to correct for problems at idle will lead to inaccurate temperatures pretty much across the board.

The cool running E2160 was simple to get repeatable test data out of but the Q6600 is going to be more of a challenge because it will be creating a lot more heat without a heatsink on it. I'm confident that I will be able to get some usable data but I need a big block of free time to do the proper testing so I'll probably have to wait a few days. At the moment, accuracy, especially for the Quad cores, is more important to me than a Minimize to Systray feature.

The dual cores I've tested all need a temp correction down low to be accurate. A Quad core is two dual cores side by side so I'd be surprised not to find the exact same thing. The problem though is trying to accurately measure for this with the amount of heat a Quad can put out. The other problem is that I might have to go to individual calibration for each core since there always seems to be 2 cores that idle at one temp and the other 2 cores idle at a different temp. There doesn't seem to be any consistency in what two cores are at the same temp. More stuff to learn but I think we're making progress. People didn't even want to consider my flaky DTS idle temp theories last year.

[gns]

Hi, I'm running a E8400 at 3.6 ghz. for some reason my core temps have a significant difference. One of the core runs at the mid 50's and the other in the mid 30's.

http://img99.imageshack.us/img99/948/realtemphx0.jpg

Should I be concerned about this. I've used other temperature reading programs and the discrepancy is still there though the individual values are different.

I've remounted the HSF and applied the thermal paste several times but there's still no change.

Anybody have any idea what the problem might be?

[Xilikon]

unclewebb, congratulations for a very nice program. I do remember our research about determining which software is accurate by using idle temp as a reference point last year... I see this is by far a great software

[Ace-a-Rue]

Hi, I'm running a E8400 at 3.6 ghz. for some reason my core temps have a significant difference. One of the core runs at the mid 50's and the other in the mid 30's.

http://img99.imageshack.us/img99/948/realtemphx0.jpg

Should I be concerned about this. I've used other temperature reading programs and the discrepancy is still there though the individual values are different.

I've remounted the HSF and applied the thermal paste several times but there's still no change.

Anybody have any idea what the problem might be?

this is a notorious problem with the Penryn and Wolfdale processors...intel did something that screwed up the DTS design in the 45NM cpu's...it is probably WHY Intel does not publish any material on DTS sensors and spec for each processor because doing so would make it easy for people to declare a warranty problem.

[coo-coo-clocker]

because doing so would make it easy for people to declare a warranty problem.

Things that make you go "Hmmmm"....

@Intel: you guys care to chime in here? Surely the [XS] population has earned the respect to get a clear scientifically-based understanding of this, right?

[unclewebb]

gns: Your problem is common with E8x00 processors. The data coming out of the DTS sensors is screwy so no software can properly interpret what the temperature is. If you believe one of your sensors is more or less accurate and neither sensor is getting stuck then it is sometimes possible to write a correction factor to line the sensors up a little better.

Here's the before and after for Brama's E8400 that I helped him with:

http://www.xtremesystems.org/forums/...&postcount=442

http://www.xtremesystems.org/forums/...&postcount=446

PM me a copy of your RealTempLog.txt file with some data from idle as well as full load running Prime / Orthos small FFTs or similar and I'll see if I can create a custom calibration factor for you to try and better balance your core temp readings.

Xilikon: Good to hear from you again. After seeing how out to lunch my temps for my E8400 were I thought what the hell, I might as well write my own temp software. It can't be any worse than what's presently available!

[flexy]

RealTemp 2.1 is reporting almost the same temps in idle as CoreTemp for my Q6600, after the ++ calibration (1C difference on some cores). Load temps are 4C lower on each core. So working fine for me I guess, keep up the good job.

yes thats what i see too!

[flexy]

It gets worse flexy. All of my testing has only been with the dual cores so far. It's likely that my 3C-5C above ambient guideline will need to be boosted up a couple of degrees to be relevant for Quad cores. I have a G0 Q6600 in stock awaiting a thorough work over so be patient. RealTemp is only a month old so with more time and more testing by myself and other users it can only get more accurate. Post what ever you learn about your Quad.

If people are upset by programmers having to create their own test methods and guess then they should write to Intel and tell them that hiding all details about the digital thermal sensors and TjMax for their processors is wrong. I don't understand what the big secret is all about.

Well that's ok. I dont understand either why we dont have that info.

Is there a way to find out real TJmax by observing when the CPU starts to throttle? <-- just answered my own question, yes of course. But the problem is not finding out TJmax (just look when PROCHOT is triggered( - but rather to display the "real" temp of the core, right?

Dont get me wrong, great program. But i have a fear that some/many people will use this and it will introduce the problem that some people will use program X (like coretemp), others realtemp.....and it creates confusion and false temp comparisons due to the different way it interprets temps.

I had similar problems reading temps from my 8800 GTS, some programs read right off the temp diode, some read what the drivers tell them. The temop difference is a whopping 10C degree !
The problem now is that people often just mention temps, but rarely what programs they use, respective what method of measuring. We dont have a reference!

I see the problem is on Intel's side, so measuring CPU temps has to be a guessing game, no matter what....but it just doesnt make it easier now with a new program which reports temps 10C degs less or requiring proper calibration first. Anyway looking forward to your finding with the G0.

[uje]

hello hello all,

i'm a little frenchy who follow that thread for a while, trying to understand my best; first of all, thanx to unclewebb for creating this software, and getting it better everyday !

i'm an owner too of a e8400, with troubles with the termal tools, so i was glad to find this out !
And finally i decided to calibrate it as you described on your first post,

so in idle at stock i got 32&#176; on 1.21 vcore,
so i got my fsb to 205 with a 6 coef at 0.976 vcore and i stayded at 31 degrees ( still idle )....., gotta say that ambiant temperature was 19 degrees

My only question is to know why am i not getting lower temps .... and so if i'm able to calibrate with a - or -- my realtemp soft ?

i tried to explain as i could thanx for reading and trying to understand..and for your next answers ! TY

Uje

[unclewebb]

uje: The (--) setting in RealTemp was designed based on what I was seeing with my E8400. The 65nm chips that I have tested seem to need a (+) or (++) setting for accurate idle temperatures.

Intel did not design the on chip digital thermal sensors (DTS) to be accurate at low temperatures so you shouldn't be surprised that your E8400 isn't accurate at idle. The DTS sensors are designed and calibrated by Intel for accurate thermal throttling at TjMax and also for thermal shut down.

The Idle Calibration feature in RealTemp helps get your idle temperatures closer to the real core temperature but it is only an estimation and is not 100&#37; exact.

In the near future I will try to re-test my E8400 at idle and I'll post some pictures of IR thermometer readings vs RealTemp readings.

flexy: I wrote RealTemp because none of the other temperature monitoring programs are considering the errors in the DTS at low idle temperatures that I have explained with my chart in post#1. I also wasn't impressed with programs getting the TjMax wrong for my E8400 and they are still doing that. When combined, CoreTemp and Everest are misreporting my core temperatures by 15C at idle. If I can continue to improve as well as prove the accuracy of RealTemp and if I can add more features then maybe 6 months from now it will become users first choice for Intel Core temperature measurement. If you like it, spread the word. If I've screwed something up, let me know so I can test it and fix it.

[Darth Vader]

unclewebb
can you add the logitech g15 support?

[Infa]

Not to be like that, but its a super program, but how is it goin with minimize to tray? only thing missing for my part hehe... great work uncle!

[unclewebb]

.... but how is it goin with minimize to tray?

Uncle is on holidays this week. Besides, my computer has been taken over by a pre-teen with a Test Drive Unlimited addiction.

Darth Vader: I probably won't be adding Logitech g15 support.

[Infa]

hehe ok, easter holiday ...enjoy it!

will be looking forward to you work later...

sayonara!

[Ced-G60]

My E8500 is reporting 42&#176;c on both core on idle (400x9.5) with an room at 19&#176;c and on watercooling...

I did the core temp test and i get 3 and 7

[Unrealer]

If you want an accurate core temperature reading then leave TjMax=85C for your processor and set the Idle Calibration to ++. You need to read the documentation for RealTemp located in post #1.

Done.

E4300 @ 3.2Ghz (1.42v full) + Big Typhoon VX @ 1800rpms
18.5&#186; ambient temp.

47&#186; @ Real Temp.
62&#186; @ Core Temp.




47&#186; @ full ??