unclewebb

Why is that wrong? Based on what? You don't list any details like your ambient conditions.

I had this argument with another Q6600 B3 owner and after he dropped his processor down to low MHz and low volts, it was obvious that TjMax=85C was correct for his B3. A B3 quad is two B2 Core2Duo processors side by side which both have a TjMax=85C. You need to do some testing before you can jump to any conclusions about what program is closer to the truth.

GigaByte

If you go through the guide at TomsHardware you will find out if you Tcase is calibrated or not, a calibrated Tcase should be 0.5 - 1c above ambient which mine is in the pics shown to you. You say using ++ idle calibration will "fix" my idle temps, it does not change my temps, it changes the displayed number and certainly does not change my load temps that have cores 5c below Tcase. So in reality its really forcing the program to say my idles are higher than they really are. While the idle calibration MAY be good for Wolfdales (as there is NO valid proof from Intel) it is not for 65nms.

Iam pretty much out of things to say or show, you have all the info you need. 85c Tj max is only on L2 and B2 steppings, but if every processor is 85c Tj max to you (even QUADS...) then thats your decision.. its your software so do what you want. I just feel people need to know that it is incorrect to have all those other steppings at 85c Tj max so they won't be so alarmed when they see their abnormal low temps with idle AND load. I think selectable Tj max will only fuel the Tj max battle but thats just my opinion on it. Iam done.. for now anyway.. when people ask about their temps I will post my side of it, feel free to post yours as well.

__________________

unclewebb

Some of the stuff posted by Computronix is simply not true:

Thermal shut down does not occur at TjMax. The Mobile Core2Duo processors that have an Intel documented TjMax=100C also have a documented shut down temperature of 125C. The desktop processors are not documented but they too will continue to run at temperatures well beyond TjMax.

Going through the guide at Tom's Hardware does not guarantee that your reported TCase temperatures exactly track the actual TCase temperatures across the entire operating range from idle to TjMax.

The quoted difference between TCase and core temperature at an idle steady state and at load both go against Intel's own in house testing of this. There are too many discrepancies in Computronix findings.

All programs agree that TjMax=85C for the original Conroe E6x00 series like my E6400. I have measured my E6400 and that seems to be 100% true. Only problem is that at idle, my E6400 will report temperatures below ambient which is exactly what your processor does when you use a TjMax=85C value.

Here's my explanation of what I've measured:



You're saying that with your processor you can set TjMax=95C and you will have accurate temps from idle to TjMax. I have now tried three completely different core processors, 45nm and 65nm, and none of them have DTS data that is changing at the exact same rate that the core temperature is changing at. They all follow that graph above where the farther you move away from TjMax, the more the measured temperature diverges from the reported core temperature.

From Anandtech:
Moreover, Intel representatives have informed us that these as-of-yet unpublished Tjunction values may actually vary from model to model - sometimes even between different steppings - and that the temperature response curves may not be entirely accurate across the whole reporting range.

That is exactly what I have measured and that is exactly what the Idle Calibration feature in RealTemp was designed to correct. By using a perfectly linear model from Idle to TjMax, all the other core temperature monitoring programs like CoreTemp, SpeedFan and Everest are all ignoring the response curve of the DTS.

All I can say is keep an open mind and try to find a friend with an accurate IR thermometer. RealTemp was designed to explain what I have measured. No other software gives me accurate core temps from idle to TjMax. Users that disagree with my interpretation of what the DTS sensor does at idle can still use RealTemp and set TjMax to whatever they believe is true. Adding
TjMax=2

to the RealTemp.ini file will boost TjMax two 5 degree notches from 85C to 95C and if you believe your B3 is TjMax=100C and not TjMax=85C then use
TjMax=3

to boost it by three notches or 15C.

Gigabyte: Why not set TjMax=85C for your processor and then set the Idle Calibration to (++) and show us your idle temps. They should be very close to your actual core temperature and at low MHz and low voltage, should be a few degrees above your ambient temperature. Make sure your ambient temperature is accurate. A home thermostat reading does not always equal the air temperature in a different room, especially at a different height.

Intel Core Temp Info:
http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf

More Additional Info:
http://www.xtremesystems.org/forums/...&postcount=319

TheSubtleKnife

hey, sorry guys for not posting lately in here...school is killer, but now since im on spring break...yay.
I should be posting some more ideas around in here soon!

bump anyways

__________________

unclewebb

Don't worry TheSubtleKnife. I've been working hard today gathering data.

Here is my latest guinea pig, an E2160 Revision M0.



Intel has a habit of creating one processor model and then making some slight changes to the basic design so they can create additional models for different price points without having to do a total redesign. The E4x00 series with 2MB of cache gets half of that disabled and then becomes processors like my E2160 with 1MB cache. They then cut the cache in half again and create things like the E1200 M0 which has only 512KB of cache. They all have the same CPUID. Some people believe that this is a way for Intel to sell defective processors. If the cache has a defect, they can disable half or three-quarters of it and still sell it as a lower end model and make some money.

So my theory is that all M0 processors start life as the same basic processor with the same TjMax. As they go down the assembly line, the multiplier and VID and cache are adjusted which creates a variety of models. These different models get stamped and shoved into the appropriate box but beneath the skin they are all quite similar.

GigaByte has brought up some troubling findings that need to be addressed. His M0 processor when using RealTemp's new TjMax=85C is reporting below ambient core temperatures at idle which everyone knows is impossible. When this same situation originally happened with the L2 - E4300 series, the solution CoreTemp came up with was to bump TjMax from 85C to 100C. Changing the TjMax will cure below ambient idle temps but it also screws up your load temps by +15C if the TjMax really isn't 100C. Not good.

The goal of RealTemp was to convert the on chip DTS data to accurate core temperatures. When I first got my E2160 installed and ran RealTemp v2.11, it was immediately obvious that TjMax=95C was wrong for my new processor. I quickly changed that to TjMax=85C for RealTemp v2.2 based on my IR thermometer testing.

With GigaByte strongly objecting to this change in TjMax, I decided I needed to re-do my E2160 testing. I recorded the IR thermometer reading and then immediately after I took a picture of the screen to compare it to what RealTemp was reporting. I'll start with a lot of pictures taken at different temperatures to compare IR gun temps to what RealTemp reports. I have cropped the original pictures down to size to make them easier to upload. I have also kept the original pictures which are all time stamped showing that the video screen shot was generally taken 2 or 3 seconds after the IR picture. It took me a couple of seconds to reposition the camera, refocus and shoot. If anyone is suspicious that my image gathering technique was done to make RealTemp look better then I have the originals that I can e-mail to you so you can check the file creation time stamp on them.

Anyway, here goes. In each group you have the IR measured temp followed by the RealTemp reported temp. Idle Calibration was set to (+) but that feature doesn't effect temperatures in this range.

IR Temp = 50.6C

RealTemp = 50C


IR Temp = 55.4C

RealTemp = 54C / 55C


IR Temp = 65.8C

RealTemp = 67C


IR Temp = 70.6C

RealTemp = 70C / 72C


I think everyone will agree that there is a very strong correlation between IR thermometer measured temperatures and what RealTemp is reporting for core temperatures. The upper range looks good and shows that TjMax=85C for my M0 processor looks like a wise choice.

Now it's time to look at the low end using the same TjMax=85C. With the Idle Calibration set to (+), you'll also see how that effects reported temperatures. I've included pictures of SpeedFan since it uses TjMax=85C for the M0 processors but doesn't have RealTemp's Idle feature so you can compare.

IR Temp = 27.8C

RealTemp = 27C SpeedFan = 25C CPU = 40C


On this processor SpeedFan, without the Idle Calibration feature, is reporting core temperatures that are a little too low. This is common on the original B2 processors, definitely the L2 processors like the E4300 and now also on the M0 processors. The other thing to note is the CPU temperature which is from a diode on the motherboard. On this Asus board, it has nothing to do with the actual TCase temperature of this processor and is completely wrong.

IR Temp = 31.8C

RealTemp = 32C SpeedFan = 29C


IR Temp = 34.0C

RealTemp = 34C SpeedFan = 32C


IR Temp = 41.0C

RealTemp = 40C SpeedFan = 39C


Once again there is a strong correlation between IR temperatures and what RealTemp is reporting. A properly chosen Idle Correction factor is also helping to make these reported temps spot on.

Everything is looking good so far but now it's time to investigate if my idle temperatures are being reported below the ambient temperature like GigaByte has found.

For this test I've reset the Idle Calibration to 0. I turned off my computer in my very cool basement and let it sit. It was left in Stand By mode for a few hours so when I turned it back on I could immediately see the reported core temperature before it had a chance to warm up too much.

I had a notepad sitting beside the computer which I used to record a room temperature of 11.2C.


Here it is. With no Idle correction factor, RealTemp is reporting 8C for the core temperature now.


RealTemp, uncorrected, is reporting this M0 E2160 3 degrees below the ambient temperature. This is exactly the same as what you discovered GigaByte. You were operating your E4500 at 0.90 volts vs 1.08 volts during my testing. Heat output of a processor is proportional to the square of voltage so your CPU was putting out approximately 70% of the heat energy as my processor and one of your cores was reporting that it was at 4C below your ambient temperature.

GigaByte, can you now see how the DTS is not linear. Without an Idle correction factor, these CPUs will report idle temperatures that are too low and impossible. That does not imply that TjMax is wrong. My testing makes it perfectly clear that TjMax=85C is definitely right for my M0 processor and our combined testing shows that our processors are behaving identically by both reporting below ambient idle temperatures.

Your correction to adjust the TjMax upward 10C to 95C will only make all of your load temperatures wrong. For accurate reported temperatures that cover the entire operating range of a processor, you need an Idle Calibration correction factor.

Down at this temperature, RealTemp uses a correction factor of 4 degrees which makes the reported idle temperatures much more accurate.



A 4 degree correction would bring the previous 8C uncorrected reading up to 12C which is above my ambient temperature of 11C. The RealTemp reading may not be 100% correct but it is definitely very close to the actual core temperature of this processor. Closer than any other core tempreature software available. RealTemp, when properly calibrated, is very accurate from idle to TjMax and beyond.

GigaByte

IR gun is not getting DTS temp its getting Tcase or even a little cooler than that. Tcase is always lower than Tjunction, now that you are viewing Tjunction with 85c Tj max its getting shown lower than normal. Heat orignates in the cores right where the DTS is, thats why they are ALWAYS hotter. If there was a way for an IR gun to get a temp reading right where the DTS is then it would be more accurate, but we all know we can't do that. The heat from the cores is speading throughout the die, then into the IHS then to your IR gun. Do you really think by the time the heat reaches the IR gun its going to be the same as the cores?

This is where and why the 5c (+/-3c) delta comes in, if I was to measure the temp half a meter above a fire it will sure be cooler than measuring the temp inside the fire assuming if my hand and my measuring device didn't melt.

__________________

opt33

According to intel documents, at idle, even at temps up to 93% of tjmax, temp of cores = tdiode/tcase, the gradient is less than 0.1C, ie near zero. To deny this assumes you know more than the intel engineers that designed and thoroughly tested the chips under very controlled conditions. See figure 5 or attached pic. In fact nearly 1/3 of programs have gradient less than 1C. Only when running stress programs like TAT does the 5C max gradient occur.
http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf

The temp gradient across the IHS, and die attach adhesive (tim1) is less than 0.2C, even at 68C, at low load steady state.
http://www.flomerics.com/flotherm/te...apers/t324.pdf

I think your issue is that you believe your diode temp to be correct and very accurate (though intel does not, hence they abandoned them), and therefore you are assuming tjmax of 85 is incorrect.

There are two possibilities, 1) your chip has a different tjmax than unclewebbs, or more likely 2) your diode chip is reading incorrectly, despite your calibration. DTS sensors are read directly from the chip, and not prone to error from interpretation. Diodes are not read directly, a mobo chip interprets electrical signals from the diode and the readings are subject to all kinds of error, including incorrect calibration in bios (regardless of your calibration), signal noise, etc. Also like DTS sensors, often the diodes may only be accurate in 60-100C range, and read way high or low in ambient range (nonlinear scaling), and thus calibrating cpu diode at ambient can induce a 5-10C error on top of other error. Intel lists these issues as one of their main reasons for abandoning diode temps in favor of DTS, in addition to the hot spot/5C gradient UNDER TAT LIKE/ENCODING type loads.

The only way you will be 100% sure whether your chip has an inaccurate diode versus different tjmax, is to measure same way unclewebb did, but I would bet on your diode is not reading accurately.

You are assuming calibrating a diode is a science. It is a guessing game, one I play myself, but would quickly reject if conflicted with more accurate core reading based on IHS calibration as above. The only science is measuring Tcase at idle steady state, undervolted and underclocked, and having intel documents back that up as valid for measuring core temps.

Also, if your chip has both a diode and a DTS, then above document tells you your diode to DTS delta should be 5C at full TAT load, and no more. If it were me, I would calibrate CPU based on that, and if it does not look ok at ambients, assume it is nonlinear, but the curiosity would get to me, I would have to measure myself.

Attached Thumbnails

 

unclewebb

opt33: Thank-you for saving me a lot of writing. I have no idea why people continue to bring up TCase vs core temperature gradient arguments when that goes completely against the controlled testing done by Intel engineers using the best of equipment. People that don't read, don't understand or choose to ignore the published data by Intel are not basing their arguments on the facts.

Thanks GigaByte for forcing me to fully investigate my E2160. It will be much easier now for other users who see this to come to their own conclusions about TjMax.

GigaByte

You had to use an idle offset to get the temps to match, doing that already shows there is a problem here. On a system with a proper Tj max being read there will be no problems with core temp being below ambient, simply forcing a temp monitoring program to show a higher number does not fix it. Its showing like that in the first place because the Tj max its using is incorrect. The only time when the IR gun can be used as 100% accurate proof is when you measure the temp while RealTemp shows 8c while your ambient is 11c.

opt33, I am not arguing against the documents, as they are for mobile processors and as we can see alot of info in there does NOT apply to desktop versions, like the Tj max on mobiles.

This pic that was taken from the 2nd lin perfectly shows what I am trying to say, the yelllow - red spot perfectly represents the heat orignating from the cores, measuring anywhere else will get you a lower temp as a delta will be in effect:

__________________

unclewebb

The problem is that you can not use a single TjMax value and have accurate core temperatures from Idle to TjMax. Why? Because the data coming from the DTS is not changing at the exact same rate that core temperatures are changing at. Intel has confirmed this and the testing I've done certainly confirms this. As soon as you move approximately 35 degrees away from TjMax it becomes very obvious that the calibration Intel did to the DTS at TjMax is no longer in effect.

The Idle Calibration feature is a band-aid, I agree. You can not "fix" a DTS sensor that puts out data that is not linear but you can adjust the readings so they better represent the real core temperature. Trying to fix this problem by adjusting TjMax higher than it actually is will lead to far more errors in reported temperatures across the entire operating range.

But you are. Intel found that Core processors at idle, where I do my testing, do not have the temperature gradients that your picture would lead people to believe and the delta between TCase and core temperature is next to nothing at an idle steady state. It's well within the +/- 1C accuracy of the Fluke IR thermometer I'm using as well as within the accuracy of the DTS itself.