Real Temp - New temp program for Intel Core processors

[unclewebb]

If these sensors have some error at TjMax and that error can grow significantly at low temperatures, it makes it pretty much impossible to prove or disprove anything. Page 14 of their recent presentation might explain the motivation for finally releasing some information about TjMax.

Future sensor design enhancement opportunities
–Report temperatures in °C
–DTS range and slope improvements

Core 2 Duo sensors are not always accurate enough for reporting absolute temperatures but looks like Core i7 will be better. Was that bone they threw the enthusiast community an engineering document or a Core i7 sales brochure?

[BeastNotro]

Hey guys i did a fresh Vista 64-bit all updates and SP1 is installed today and i left everything at STOCK in bios, so CPU is at stock and CE1 and speed stepping is on and whatever else in Stock Bios settings.

Why i did a fresh install ... well cause i frigged some files up..... anyhoo i down clocked before i did fresh install.

Using BIOS 1306.

Here is picture of my IDLE temps at STOCK the way all is suppose to be.... not sure if this helps you out all and it is night time here 9:52 PM and it is very cool in my room right now.

How should i calibrate if need to be? as i get confused, i am serious on calibrating stuff.

[Ol'Baditude]

If I were you, I'd get a good room thermometer to make sure you have an accurate idea of your ambient room temperature (not sure what your idea of "cool" is, sry) and adjust the offsets so each core reads roughly 7 degrees above that number. I see you have good air cooling, so that offset amount should do it. They will be constantly changing slightly so you will never get it right on, but it doesn't matter. I left it for 10 minutes to make sure the readings were stable, then rebooted and clocked up.

In my case, since I am under a good water loop, I adjusted my offsets to read 4degC. above ambient at the same underclocked settings you are using. Ambient on that night happened to be 22degC.
My overclocked idle temperatures now read roughly 5 degrees above ambient room temperature now. Delta CPU temp from idle to load ~12Deg.

[BeastNotro]

Here is (Picture Below) of it at OC idle temp and again it is very cool in here this morning 8:25 am .. i checked before and Ambient is usually around 20-24C througout the day, but if she gets really hot out in the day it can be 28C Ambient.

I don't have a thermometer, but i borrowed a Lazor Temp Probe about 3-4 weeks ago and checked on a fairly warm day and it was 24C in my room..

So should i just adjust all cores to 30C? or now thats its at OC 3.6ghz ... adjust them to 37C?

This is IDLE right now (pic Below) ... not calibrated at OC speed.

[Ol'Baditude]

I don't think you want to be doing your offset while you're clocked. I've tried that(saved my underclocked offsets and tested with OC'd calibration) and it made my temps read ~5deg higher on average than they would have doing a normal idle calibration. I think that goes against the whole line of thinking that webb had in mind when he developed the calibration procedure.

If your ambient is 24 deg, and you have your undeclock settings active, I would say that setting all the sensors as close to possible to 30deg would be a perfectly acceptable way to do it.

[WaterFlex]

BeastNotro
Nice temps!!! And it is 2.75! Good job.

[unclewebb]

WaterFlex: The Beast is using RealTemp 2.75 but at the moment he has decided that using TjMax=95C is closer to the truth. The Intel announcement that was going to get everyone on the same page now has some users using, TjMax=95C, some are using TjMax=97C and others are using the Intel recommended TjMax=100C. By not providing us with any test results or any information on the range that TjMax can vary by, we're now guessing more than ever. Intel's official documentation that shows TjMax varies by plus or minus X doesn't help very much.

My opinion at the moment is that TjMax=100C is just that, the absolute maximum TjMax that E8400 processors leave the factory at. The real TjMax might be something like 97.5C +/- 2.5C or somewhere between 95C and 100C. That makes the most sense to me but it's impossible to prove. Intel knows how big the margin of error X is but they're not telling because they don't agree with programs converting digital thermal sensor data to absolute temperatures. These sensors were never designed for that purpose.

I sent the Beast a super sized calibration info package so it should start making a little more sense to him. His Q6600 has nearly identical sensors to the Q6600 I'm using.

[BeastNotro]

READ ALL TEXT

Ok here is some more testing and pictures i am 3.4ghz for now until i figure out something.

Ok as per our conversations in PM's UncleWebb here is some pictures for you to analyze for me.

I did Orthos Blend Test and Prime 95 Small FTT 64-bit (plus a image of no calibration) and you will see Prime 95 Small FTT Pushes a little more.

The tests are Full Load 4 - 10 min or Orthos and or Prime 95.

Calibrated as per you PM ..

TJMax 100 is on Left and TJMAX 95 is on the right after Calibrations Per your PM.. IDLE TEMPS AFTER CALIBRATIONS.




Prime 95 64-bit version SMALL FTT - TJMax 100 is on Left and TJMAX 95 is on the right after Calibrations Per your PM and in the middle is TJMAX 95 NO CALIBRATION.




Orthos Running Blend Test - TJMax 100 is on Left and TJMAX 95 is on the right after Calibrations Per your PM and in the middle is TJMAX 95 NO CALIBRATION.

[JohnZS]

JohnZS nice temps dude

Thanks
To be honest I am surprised to

This level of measurement would never be possible were it not for all the hard work put in by Unclewebb.

Not meaning to cause a debate or anything but before I tried RealTemp I was trying to use a competitor application for temperature monitoring...it was that confusing I had no idea what was actually going on in my Quad.

Then along came real temp which had the real TjMAX AND the ability to calibrate (or shall I say account for Extreme Sensors) and now I am confident that my Core2 Quad is running at ~X degrees C

Thank you once again Unclewebb
2.75 caters for Extreme Sensors more than the previous editions.

John

[jason4207]

Beast, are you calibrating idle temps at 3.4GHz?

1st thing to do is pick a Tjmax...that's a tough call.

Next, you need to be super-low...like 200x6=1.2GHz or 266x6=1.6GHz and at ~1.05v to do the idle calibration properly. Once calibrated you can go back to your OC settings and tweak your calibration settings so that your temps track well together when put under load.

[unclewebb]

I was doing some more testing last night with an E8400 and a Q6600 - G0. At TjMax=95C, the IHS measured temp and the RealTemp reported temp are exactly the same at 80C for both of them. If I use the Intel specified TjMax=100C then the core temperature is reported 5C higher than what the IR thermometer shows.

With the processor idle and allowed to heat soak at 80C for 15, 20, 25 minutes, the 5C gradient stays the same. If this is an Intel measured gradient then I'd be interested in seeing their testing. With the known but undocumented amount of error in these sensors, it's impossible to prove what TjMax really is besides likely somewhere between 95C and 100C.

When testing an original B2 E6400, the long assumed TjMax=85C value can't be right if Intel is telling us that TjMax=100C for the E8400. There is a 10C difference between these two so the default would have to be TjMax=90C as rge mentioned before. Same goes for the M0 - E2160 that I re-tested. The Intel documented Thermal Specification for this one doesn't add up.

The next official release of RealTemp will use the higher TjMax values across the board. For the Extremes I'll have to look at the Thermal Specification. If you don't believe a 5C gradient exists then you can deduct 5C from TjMax and go back to RealTemp's original values. You could also split the difference and knock off 2C or 3C from TjMax.

I have to use the Intel released TjMax values and I have to apply that knowledge to 65nm so there will be consistency between 45nm and 65nm. Maybe at the next IDF Intel will get around to releasing some info about 65nm so software developers can do less guessing and assuming when trying to figure out TjMax for 65nm. I'll post some pics in a day or two.

[Freaky Freezer]

I'm very sorry to ask since I know the answer must be hidden somewhere amongst these 87 pages, but can someone tell me what is the "decided" or "standard" Tj. Max for the E8400's or for all the 45nm CPU's?

[rge]

Intel stated that all non-extreme 45nm have tjmax of 100 and that a few C of variance or error will exist part to part.

If you read a lot of intel documents, you will see that intel uses "worst case" or "max over life of cpu" and tests under "worst case" ambient ie 35-40C. Intel does not define tjmax, but if they did I have no doubt they would define as "maximum possible", they do not use averages for critical values. Take a look at tjunctionoffset definition or even power density from intel celeron in pic below.

If your question is what is intel using, ie what is max possible tj, it is 100.

Standard, decided or even average tj....I think as unclewebb just said between 95 and 100. I tried putting in 97.5 for realtemp, but it told me to put in an integer...so I used 97

[Merman]

I'm very sorry to ask since I know the answer must be hidden somewhere amongst these 87 pages, but can someone tell me what is the "decided" or "standard" Tj. Max for the E8400's or for all the 45nm CPU's?

Intel has stated 100C. Look at post number 2152 on page 87.

[i43]

well, if ''They..'' say 100 C , so ok then ... no problem .. i made/ run '' my own worst case scenario '' myself with my (immortal ) E8400 @4050mhz /full air cooled Rampage Extreme
so left 2 instances of RealTemp (v.2.70 on 95C TjMax and v.2.75 on 100 C TjMax ) both are NOT 'idle' CALIBRATED, though ,
i know that CPU have bad? 9 C degree idle delay/difference between core#0 & core#1 ,
but when approaching close to the '' boiling point '' ( ~ 80 ~ 85 C ),
the delay will be gone ( 'numbers' by my understanding of course, are depending on tjmax value set by monitoring program )

also leave Everest to do the monitoring graph for throttling .. BTW everst was set
on 'Automatic' so i assume it take tjmax-100 ..
there was *CPU TM Function* ENABLED in RE bios

screen 1: it has run Prime over an hour when '' my own worst case scenario '' kicks /this means - i lift off CPU fan from Noctua NH-C12P , graph show CPU temp it rising up ... to throttling point ..


screen 2: cool down / fan back on cooler , prime stopped my self



so with my setup , this the throttling point moves upper 5C degrees from -- 94C to 99C -- according to new version of RealTemp v2.75
see the '' Maximum '' field on Realtemp panels

[rge]

Yep, when you alter tjmax, you alter the temp that is displayed at DTS=0 (throttling), but of course you do not alter the actual throttling temperature, whatever that is somewhere between 95 and 100. So if you raise tjmax 5C, the displayed temp at throttling is 5C higher (though actual temp does not change). If you set tjmax to 70, your cpu would still throttle at 95-100C, but the temp displayed would be at or near 70C.

[Freaky Freezer]

Intel says 100 Tj. Max. Do the majority agree with that number? I saw it has +/- next to it and a 97c-98c. So what is the best Tj. Max to use? Or is the jury still a long ways from deciding?

[unclewebb]

i43: Now that Intel has declared that TjMax=100C for your E8400, RealTemp 2.75 and CoreTemp 0.99.3 are finally in agreement, at TjMax at least!

The only thing left to argue about is that one of your sensors suffers from "slope error" as Intel likes to call it. CoreTemp doesn't let you do anything about that. You could try doing the RealTemp calibration or you could assume that core1 is more or less right and set a Calibration factor of approximately -5.0 for core0. Adjust that number if you have to so that your core temps are equal at idle. Use the Intel approved TjMax=100C for your next test.

Try running different loads and see how your two cores track each other as the temps go up and down. Does the latest calibration formula in v2.75 allow your two cores to report similar temperatures when the load is the same on both cores? Just curious. I know how much you like to test so I thought I'd give you something new to check out.

I just want to show users that even with the same TjMax, something still needs to be done with the "slope error" many of these 45nm processors have.

Freaky Freezer: Intel has stated TjMax=100C so we have to go along with that. They've also mentioned that these sensors can show significant slope error at lower temperatures and that TjMax is not an exact number. Plus or minus a couple of degrees at TjMax might be pretty typical but Intel didn't release an official range and likely never will.

[Freaky Freezer]

unclewebb, I'm using a Tj. Max of 95c. Just to get a final word from you, you agree that Tj. Max of 100c for the E8400 is the best to use?

Also, at a current Tj. Max of 95c. My E8400 sits idle between 30-34 and 30-35, is this the "slope error" you're referring to? Or have I improperly applied thermal paste?

[unclewebb]

Freaky Freezer: For an E8400, I would suggest using the Intel recommended value of TjMax=100C. Intel hasn't released enough information about how that number is determined or what the typical range of error in that number is. For 45nm, there is likely a couple of degrees of error in that number but no easy way to determine or prove an exact TjMax value so all we can do is accept what they've told us.

If you take the time to try and calibrate RealTemp to your sensors as outlined in the documentation, any slight error in TjMax will be reduced.



"Slope error" is when two sensors at full load provide very similar temperatures but at idle they can differ significantly. If you graphed this on a piece of paper, the two temperature curves would be at different slopes.

In my example, at 70C both cores are reporting the same temperature. In a perfect world, if these sensors moved at the exact same rate as the temperature changed, they would both decrease at the exact same rate and follow that pink line in the middle. At idle they would both report the same, correct value of 35C.

In this example let's assume core0 is the yellow, top line. The sensor data changes at a slower rate than the actual temperature is changing. By the time both cores have cooled down to 35C, it is still reporting 45C which is wrong. Core1 does the opposite. It's sensor is overly eager and it changes in value faster than the core temperature is changing. When both cores have cooled down to 35C, core1 has gone crazy and is only reporting 25C. That makes for a huge difference in reported temperatures between two cores that are actually the same temperature.

My example that has two cores reporting a 20C difference might seem a little extreme but if you go back a page or two in this thread you will see that JohnZS has an Expensive processor with a difference at idle of 15C between cores. Intel has been nice enough to tell us that this problem exists but hasn't given us any approved way to compensate for this problem. RealTemp uses a formula to compensate for this and lets a user try and figure out if his cores are reading too high or too low.

Read the documentation and send me a PM if you need some more detailed help with the calibration procedure.

[kpo6969]

unclewebb,
If calibrated with 95c and it was right on would you still go with 100c?
http://www.xtremesystems.org/forums/...postcount=1740
I am using 100c now but I think 97c would be more accurate (or as accurate as can be expected)?
Thanks

[Freaky Freezer]

Thank you so much unclewebb. I really appreciate your in depth response. I won't do any calibrating because in all honesty I use Everest for all my temps and the new version of Everest only allows increments of 5 when adjusting Tj. Max. So if I find out my Tj. Max is 97-98, I wouldn't be able to enter that data into Everest anyway. I did use an earlier version of RealTemp and it worked great but I just have all my other temps under Everest so it's easier to just use that. But thanks again.

F%$#ing Intel

[unclewebb]

kpo6969: Officially, Intel says TjMax=100C so for compatibility I'm going to be recommending that. For maximum accuracy, TjMax=97C looks like a good pick for you. If you go with TjMax=100C then for better accuracy you'd probably need to use a small negative calibration factor to get better idle temps. Intel says that TjMax is not an exact number so if you need to move it around a couple of degrees then I no longer see that as being a problem.

I have my Q6600 in at the moment and it doesn't seem to have the slope error that many 65nm sensors do. When using TjMax=100C and assuming a 5C gradient between IHS and core, it seems to track pretty well.

I've gone up to about 80C so far with it and the small differences between cores is staying fairly consistent. I'm only at idle and the heatsink is just floating on top but it seems to indicate a slight difference of TjMax from core to core. I used to avoid manipulating TjMax because I assumed that it was a fixed value but adjusting it by a degree or two might be the best way to get the temps equalized on this processor. Now that Intel has said that there can be a slight variation in TjMax from one core to another, a slight manipulation in TjMax looks like a new option.

If two cores stay a small fixed amount apart from idle to your maximum temperature, I'd try to use a TjMax adjustment and if the difference between cores decreases as things heat up then I'd go with the standard RealTemp calibration adjustments to correct for "slope error."

[Freaky Freezer]

I just finished reseating my GTX waterblock on top of my E8400. First I cleaned it, purified it and applied AS5 according to arctic silver's directions for a dual core CPU. My temps are still 4-5 degrees apart between Core's 0 and 1. Anyone have any ideas?

(35C and 39C at idle with Tj. Max of 100. It is overclocked to 3.9ghz at 1.344 Vcore with EIST and and C1E enabled. Room temp is 23.5C and case temp is 40C. Are these temps good, bad or average?)

My smart Guardian reads my CPU temp at 23C-25C. What sensor is Smart Guardian reading? Tcase temp?

[DJSUB]

i just want to ask what's the correct tjmax to set in realtemp 2.75 using Q6600 processor?
keep up the good work uncleweb!

thanks in advance!