Real Temp - New temp program for Intel Core processors

Here's how I calibrate the RealTemp test equipment.
Point Fluke temp gun outside and compare.

http://img89.imageshack.us/img89/893/ocweathergk8.png

Yep, looks about right to me! :yepp:
Perfect overclocking weather in western Canada today.

Bhairava: I screwed up when I first wrote RealTemp. It was originally based on measurements of the IHS. rge did some testing and determined that the actual core temperature is likely about 5C higher than my IHS temp readings so I adjusted TJMax 5C higher between version 2.70 and version 2.90.

Reported Temperature = TJMax - Digital Thermal Sensor Reading

The DTS (Distance to TJMax) data should be the same no matter what software you use but if the assumed TJMax is different then the reported temperatures will be different. TJMax is not a fixed value. There is an unknown amount of error in this number and Intel has not given the user community any guidance on how much error at their calibration point there really is.

solarises: You list your ambient temperature as 46C - 51C. Ambient temperature is your room temperature. If it was over 46C in my room I sure as hell wouldn't be sitting there using my computer. :D

JohnZS: I always like seeing your numbers because it helps demonstrate some of my crazy theories. Between the 30% and 88% level, core2 goes from 65.9 to 49.8 which is a difference of 16.1 and core3 goes from 60.6 to 44.3 which is a difference of 16.3. The slope error difference is only 0.2 degrees. The rest of the difference is all a difference in TJMax between two cores on the same die. How's that for quality control?

Your data suggests different TJMax values for each one of your cores. Some cores are pretty close, core0 vs core1, while others like core1 vs core2 might have a TJMax difference of 10C or 11C. :eek:

You don't read about stuff like that in the publicly available Intel docs but the numbers in the Cool Down Test tell a story that is hard to argue against. I still plan to have a good look at your numbers and come up with a calibration suggestion for you so bug me on a regular basis. I should have time in the next couple of weeks.

Venturer: rge's testing showed that there is about a 25C difference between the core temperature at full load and the Tcase temperature which is at the center of the chip. A good heatsink does not get nearly as hot to the touch as one would think. I'll go get my Q6600 up over 90C and take a measurement of the top of my Tuniq. We all need some exact numbers to understand this better.

Quote:

---

Originally Posted by Bhairava

Hi all.

http://e.imagehost.org/t/0587/3800stress.jpg

Which one i should believe?

---

Ciao Bhairava!

Are you the one from hwupgrade?

We meet again here; cool! :)

Quote:

---

Originally Posted by unclewebb

Venturer: rge's testing showed that there is about a 25C difference between the core temperature at full load and the Tcase temperature which is at the center of the chip. A good heatsink does not get nearly as hot to the touch as one would think. I'll go get my Q6600 up over 90C and take a measurement of the top of my Tuniq. We all need some exact numbers to understand this better.

---

Good........i hope you will not blow your q6600. However you should implement tcase reading in realtemp.
For tcase temp value i'm using speedfan and Intel Desktop utilities.



Look at this screenshot.
http://img222.imageshack.us/img222/9...tiesak1.th.jpg

You can notice Intel Desktop utilities reports two system temps: The first Internal system temp is the Tcase temp; the same of Speedfan's "internal temp";
the second system internal temp that reads only Intel D.U i think is the sensor on the motherboard but i'm not sure.
Also notice in speedfan there is a big discrepancy between the temps of the first 2 cores an the last ones.

What do you think about these values?

Quote:

---

Originally Posted by unclewebb

Bhairava: I screwed up when I first wrote RealTemp. It was originally based on measurements of the IHS. rge did some testing and determined that the actual core temperature is likely about 5C higher than my IHS temp readings so I adjusted TJMax 5C higher between version 2.70 and version 2.90.

Reported Temperature = TJMax - Digital Thermal Sensor Reading

The DTS (Distance to TJMax) data should be the same no matter what software you use but if the assumed TJMax is different then the reported temperatures will be different. TJMax is not a fixed value. There is an unknown amount of error in this number and Intel has not given the user community any guidance on how much error at their calibration point there really is.

---

Thank you uncle, i just read your previous explanations, sorry to ask questions before having searched this thread.
Ok so it's seems that is impossible to determine the real cores temperature, right?

This is my sensor test:
http://e.imagehost.org/t/0836/cooldown_test.jpg

I don't know how to understand it. Thank you mate.

Quote:

---

Originally Posted by Venturer

Ciao Bhairava!

Are you the one from hwupgrade?

We meet again here; cool! :)

---

Ehi ciao sono io! Eheh il mondo di internet è piccolissimo. ;)

Here's how my Q6600 looks running Prime Small FFTs with the fan turned off.
I'm using a split TJMax (100 / 105) which looks believable.

http://img242.imageshack.us/img242/8070/heattesths2.png

The side of the Tuniq tower is showing 62.6C

http://img514.imageshack.us/img514/1...niqsidejm4.jpg

and the top is only at 47C which is 10C above normal body temperature of 37C.

http://img514.imageshack.us/img514/5044/tuniqtopux3.jpg

The top feels warm but not at all hot. The side feels hot but I can put my hand on it without worrying about losing any skin.

During normal operation with the Tuniq fan turned on, even when the cores are very hot, the cooler isn't that hot to the touch. With a Dual Core, the Tuniq will be even cooler to the touch since there is less heat being dissipated.

Bhairava: Your Cool Down Test shows that core0 is stuck at a Distance to TJMax of 64. Core1 might also be stuck at 64 but it's hard to tell from this one test. You would need to try and operate your CPU in a cooler environment to prove it. You could also reduce your overclock back to default settings with less core voltage and it would probably show the sticking points better.

You could also decide not to get too concerned about temps because while running full Prime load you are still 48C away from TJMax. There's no need to give your core temperature a second thought. It's also very likely that your two cores are using a different TJMax. Maybe 105/100 are your actual TJMax values. Find out the sticking points of your sensors first.

Venturer: You have a bunch of temps from a bunch of uncalibrated sensors. None of these are 100% accurate, 100% of the time. I trust TCase even less than core temperature readings and I know Intel agrees with that statement. That's the whole reason they went to core temperature sensors. For better accuracy to determine thermal throttling correctly. The core sensors have a lot of issues but they do a pretty good job at what they're designed to do; thermal throttling and thermal shut down.

uncle, can you point that gun on a heatpipe? I'm curious what temperature it have.

Uncle; Burebista do you think changing my heatsink with Thermalright Ultima90 plus papst 92mm PWM fan 3500rpm should improve my situation or will it be useless?

Thank you!

Man I dunno what to say. :(
I'm puzzled by your 18°C drop with case open so I hesitate to say "Yes TR Ultima is your ultimate solution".

Quote:

---

Originally Posted by burebista

uncle, can you point that gun on a heatpipe?

---

An IR thermometer doesn't work very accurately when pointed at a small, curved shiny heat pipe. Based on the temps of the fins I'd guess somewhere around 70C for the heat pipes. With the board inside a case it's hard to maneuver around it for accurate temps.

Venturer: A new cooler probably won't get you very much extra real world performance. I'd save your money for future upgrades. The QX6800 was a wonderful processor when it came out but now, for $320 bucks, you can get a Q9450 Yorkie that will run faster or a lot cooler at the same MHz. It also has a higher TJMax so you can run it hotter before it'll start to thermal throttle.

I'd spend my time learning to kill unnecessary background processes that create heat and do what you can to improve case temperatures.

Quote:

---

Originally Posted by unclewebb

JohnZS: I always like seeing your numbers because it helps demonstrate some of my crazy theories.

Your data suggests different TJMax values for each one of your cores.

---

And another one on ocforums. :)

Alright Unclewebb.
Going to ask for your input here. I hope I come across correctly this time.

This is not a "gotcha" question. It is what it is. I LOVE your program. It has helped me feel very comfortable overclocking my Q6600 on my stock cooler.

But therin lies my question. RealTemp is a great tool. But for someone like me on stock air from intel....I have noticed something and would like your input.

Ok, realtemp, test sensors, small FFT's......heat's the $%*&^ out of cores as it should. I'm at 400 * 8 for 3.2ghz and just finished running your sensor test.
I was 16 C from hitting TJmax @ 100%. As I said, it's a great tool. That test told me some things: A) Vcore needs to be lower, OR, (duh) I need better cooling, and B) Probably not going to go over that 3.2 ghz speed until I have accomplished lower vcore AND better cooling.

But I just realized, that's not necessarily all the info I need. I think a fairly "standardized" Overclocking test for stability comes in the form of Prime95 running a BLEND test, usualy overnight, sometimes longer.

What I have noticed, when running blend test, my distance to TJmax doesn't even come close the 16 c I just saw. It MIGHT get up to 30 c to Tjmax or 28c.

So I guess my question is, putting 2 and 2 together, don't you think I could try to push to 3.3ghz, 3.4 ghz. maybe even 3.6.....run the prime blend test for stability, but definately stay the hell away from your small fft test?

The Blend stability test runs cooler than your sensor test, and I believe the blend test * IF IT PASSES * , puts as much stress on your compenents than any real world application. such as video encoders, etc....

Yes, Blend test is more real-life orientated because it stress CPU-MEM-NB like in daily use.
Small FTTs are for heating up your CPU but putting much less stress on MEM and NB.
Linpack (LinX/IBT) is for testing your cooler and CPU stability, it heat up your CPU like no tommorow. :)

concretefire: I agree with you 100%. I think Prime Small FFTs stable for 101 hours is over kill for a Quad core processor. It doesn't represent the typical load that most people run on their Quads. Linpack (LinX/IBT) is another order of magnitude on the over kill scale.

You know best what applications you run on your computer and how important your results are. If you're Homer Simpson and running a nuclear power plant then you need a little more head room and stability compared to shooting bad guys in the latest game.

The bottom line is stability in the applications you normally run. If you're stable and your CPU is not thermal throttling then you don't have to worry about your core temperature. I ran my E8400 at an average of 98C with the CPU fan off for 3 hours a while ago just to prove that point. It was Prime Small FFTs stable and didn't catch on fire or blow up.

When overclocking with the Intel OEM cooler, my E6400 always told me when I was pushing it too hard. I would lose stability as the core temperature got up near 70C. If you don't overclock them too hard, you can run a Core based CPU reliably at a higher temperature. My E6400 was very happy at 3200 MHz, I used to run it mostly at 3400 MHz with the OEM cooler and if I dragged my computer to the basement it was Prime Small FFTs stable for 4+ hours at a hair over 3500 MHz. I may not have been Prime stable upstairs at 3500 MHz but it was game stable at that speed.

I use Small FFTs during my test because it is the only program I know of that provides a decent amount of heat. The power consumption at the wall is very consistent which insures that the CPU heat output is also very consistent. I need those characteristics for consistent, repeatable results during the CPU Cool Down Test.

http://i82.photobucket.com/albums/j2...g?t=1229377129
Core 1 is stuck?

Before using realtemp2.90 i already thought the sensor were stuck @core 1.
It never shows temps below 36 degrees, even on stock speeds.

Got my cpu watercooled.

Hey everyone, long time listener, first time caller.

I have a couple/few question that I hope you guys can answer for me. I'm wondering why Intel states that for my chip (b3 q6600) tcase max temperature is 62c. Then they release info saying that tjunction max is (excuse me I don't remember exactly what it was) around 80c. Doesn't that mean by reaching their specified tjunction max, we would be by a large margin, over their specified tcase temperature? Seems like a contradiction, or just messed up info to me.:confused:

Also I have been debating with the people over at Tom's Hardware about operation temperatures and their relationship with the life expectancy of the processor. If I were to run my q6600 at let's say 80c with a voltage setting of 1.4, is it safe to assume that no damage is occurring provided the voltage stays under max specs? Or should I expect my chip to last about as long as the Intel warranty of three years, or less even? :eek:

I'm also curious if any of you have done any prolonged high temperature testing, to see if any damage has been sustained.

Thanks to all in advance

Thanks Unclewebb. This whole overclocking thing is like an adventure really. You never know what the next setting might reveal. Good or bad. lol.

I do have a request though for your program. A snapshot feature. (screenshot) I would like to see it on the cool down page. After the cool down goes all the way down to idle, a "greyed out" box labled "screenshot", would then become un-greyed, and you take a snapshot.

I have done this with vista's sniping tool, and it's great. For each step along my overclocking journey, I have created a folder such as "Default", 2.5, 2.6, 2.7,2.8, etc...each with thier own folder representing my CPU overclock. Then I run benches, tests, everest, realtemp, etc....and put all the data in each folder.

Just a thought. Keep up the great work.

Bullu$: Looks like RealTemp helped you confirm what you already suspected. Core1 is stuck at a Distance to TJMax of 64. When the load drops from 29.1% to 1.0%, core0 cools down 2.6 degrees. That's normal. Core0 doesn't budge. That's not normal. That looks to me like a stuck sensor. There is also another indication when you compare the 38.4% level to the 29.1% level. Core0 moves 3.1 while core1 only moves 1.0. I'm a little biased but I love this new test. It makes it very simple to spot a sensor or sensors that have issues.

SportsFanBoy: According to Intel, TCase is measured with a thermocouple attached to the geometric center of the IHS by cutting a groove into the top of your CPU. When you are running an application like Prime95 at full load, the core temperature might be 25C hotter than the TCase temperature. These CPUs dissipate heat that quickly over that small of a distance. Contact rge who did some very interesting testing of this or scroll back about 20 pages or so to find his work. An 80C core temperature would not put your TCase temperature over Intel's rating.

Intel refers to their recent release as TJ Target numbers. A TJ Target number of 80C may translate into an actual TJMax of closer to 90C. The information released by Intel last summer and in the fall didn't really clear this up. The TJ Target numbers do not equal TJ Max for a significant number of 65nm CPUs that they've produced.

I have no idea how much long term heat an Intel CPU can take. I do know that they seem to be able to take far more heat than people would think. Here's an example you can share with your friends over at Tom's. Three hours running Prime95 Small FFTs without the CPU fan turned on definitely created some heat: :D

http://img204.imageshack.us/img204/276/hote8400fw5.png

If your Q6600 is running stable and not thermal throttling then don't worry about the heat. If Intel thought 80C was too much heat then they'd lower TJMax and prevent their CPUs from ever reaching that temperature. Intel actually did the opposite and raised TJMax as their manufacturing process matured. Heat related warranty / RMA claims must have been a non issue. I've never had a problem even with all of my insane testing.

concretefire: I'll think about your snapshot idea for RealTemp. I just use the Print Screen key on my keyboard. After you push this you can go into any image editing program and Paste your screen shot from the Clipboard in and save it to the file type of your choice. I use an old version of Photoshop Elements 2.0 but even Paint is good enough for screen shots. Irfanview is an excellent freeware image display program that also lets you Paste and Crop images without a lot of effort. If you want an image of only RealTemp then hold down the ALT key when you push Print Screen. This will send an image of only RealTemp to your Clipboard.

Thanks for the quick reply Uncle, maybe I should go back and read some of the previous pages, thanks agian

I'm not sure if RealTemp is designed for mobile CPUs, but I found that the latest (2.90) is having issues grabbing and calculating the FSB and actual clock of my P8400 (Centrino 2):
http://dl.getdropbox.com/u/2028/RealTempFSBBug.png
Just checked, now my FSB is at 42.45 :sofa: Maybe the Speedstep is messing it up? I've noticed it takes around 2 minutes before RealTemp tried to show the clock speed and FSB.

Currently using 2.90 and i love the new interface layout, the only thing missing ( for me ) is a real time vCore read out.
Thanks for this wonderful update unclewebb.

Metric: RealTemp does not presently calculate the MHz correctly for mobile chips when idle with C1E / Speedstep engaged. It's definitely messed up. :)

Edit: Maybe it's not until C2E or higher get engaged that it starts to get confused.

In your screen shot the x 6.0 multiplier is correct but the MHz are being averaged. If your bus speed is 300 MHz and your multiplier is 6.0 then in theory it should report 1800 MHz. What RealTemp does is if you are idle and your cores are asleep for half of the sampling period then instead of reporting 1800 MHz, it would report 900 MHz. For some users this is a feature while for most users, it's a bug. That's also why RealTemp might report nothing. It's looking for a consistent MHz value but it can't find one because the average MHz is going up and down depending on the load. I'll try to come up with an option / fix so you can see 1800 MHz instead of watching the MHz jump all over the place or not reported at all.

Try running a program like Prime95 to load your cores for 30 seconds. Does it show the correct MHz and multiplier then?

Your welcome Demo.

Quote:

---

Originally Posted by unclewebb

Bullu$: Looks like RealTemp helped you confirm what you already suspected. Core1 is stuck at a Distance to TJMax of 64. When the load drops from 29.1% to 1.0%, core0 cools down 2.6 degrees. That's normal. Core0 doesn't budge. That's not normal. That looks to me like a stuck sensor. There is also another indication when you compare the 38.4% level to the 29.1% level. Core0 moves 3.1 while core1 only moves 1.0. I'm a little biased but I love this new test. It makes it very simple to spot a sensor or sensors that have issues.
.....

---

Yeah,u made a nice program:up:

Quote:

---

Core0 doesn't budge. That's not normal. That looks to me like a stuck sensor.

---

But ,what i understand is that both my cores are stuck?:confused:

Bullu$: Core1 is stuck at a Distance to TJMax of 64. Core0 will also get stuck at some point but during your test it shows that it is still moving at 65 and can read at least to 66. You would have to lower your core voltage and MHz and room temperature to see how far this sensor can go before it becomes stuck.

This doesn't work with older processors? I have an 805 D and it says it is not compatible. The proc is not that old...