Real Temp - New temp program for Intel Core processors

[shazza]

Agree ... it's working fantastic for me.

Here's my latest data after the cool down test. Running an i7 920 @ 3.8 on the EVGA X58 board. Without this program, I'd have a lot less confidence experimenting with overclocking. I don't freak out because the cores read different temps, as I can check their responsiveness using the Cool Down Test. It's also a quick and easy stability check before moving to longer tests.

Just wanted to say a big to unclewebb and everyone who has worked to make this a great application.

[unclewebb]

Thanks for the support. I find when this thread gets quiet that means everything must be working good. No complaints is usually a good thing.

The latest version has been polished up a little and should be ready at the main TechPowerUp site in a day or two. Stay tuned.

[randomizer]

The latest version has been polished up a little and should be ready at the main TechPowerUp site in a day or two. Stay tuned.

Excellent! No more people downloading obsolete versions

[Snafu]

@unclewebb

Any suggestions on how to unstick a sensor if the sensors are faulty?

My core temps are reading significantly different values and wondered if this is similar to what others may be experiencing and hopefully found a fix. I am dying for a fix because I don't feel comfortable overclocking this system until I know what are the real temps.

My system is basically:

* Intel E8400
* Gigabyte EP35-DS3P
* OCZ PC2-9200 Reapers
* Seagate 3GB/s HDD

The cpu is water cooled with 120x240 and 120x120 rads.

I flashed to the latest bios (F6a) and found that the temperatures were pretty much out of whack.

I tried the calibration approach used for Real Temp by dropping the core speed to 1.6GHz (6x266) and core voltage to 1.10v. The temp readings were 46/20 at idle (water cooled and RT at 18C).

Reset the core speed to normal 3.0GHz (9x333) and core voltage set to “normal”. Idle temps were the same (45/20) as the core running at 1.6GHz. The load temps were 51/33 as seen in Real Temp after running Orthos.

So the bios update did not fix the temperature readings and trying to recalibrate Real Temp won’t work to try and find semi-accurate temp readings.

Are there any methods to unstick a sensor?

One other thought is that everything is working normal and one core is getting hot due to a bad contact with the heat shield. The only solution in this case would be to pull it off. Since this would be my first attempt so I would prefer to try other approaches first.

Feedback is appreciated. Thanks

[Richard Dower]

Any new beta gonna be released?

[randomizer]

@Snafu: Run the CPU cooldown test in Real Temp and post your results so we can see how your sensors respond across a range of load

[unclewebb]

Snafu: There's no way to unstick a sticking sensor. The sticking point of the vast majority of sensors never changes during the lifetime of the processor.

Don't let this hold you back from trying to overclock and get some MHz out of your CPU. The majority of sensors will start to move and register more or less correctly once the temperature is greater than the sticking point. You can't do a proper calibration with a stuck sensor but you might be able to improve its accuracy at higher temperatures when they start moving.

randomizer is right. The CPU Cool Dow Test can tell me a lot of information about your CPU and its sensors. All it takes is about 10 minutes of your time. I get a little busy with programming project RealTemp sometimes so if you don't hear back from me just send me PM messages and e-mails until my in-box is overflowing!

[Snafu]

Thanks randomizer and unclewebb. The cool down test is being run and I will post the results. I was running v2.70 but have downloaded the beta.

BTW love the program. Very runs great, nice and compact. Excellent!

[Snafu]

Here are the results of running the cooldown test (PM on the way):

[unclewebb]

Snafu: The good news is that neither of your sensors are sticking in the temperature range that you've done your test. Typically when there is a large difference at idle it's because one of your sensors has become stuck but that is not the case here.

The problem you have is two sensors with totally different temperature slopes. If you compare the Idle level to the 87.6% level, core 0 changes by 5.3 while core 1 changes by 10.0. Intel says that the slope of a sensor might vary by 10% so if you compare two of them you might see a 20% difference but here the combined difference is closer to 100%. That's not good and is likely the main reason why your temps look so screwy.

If core 0 had an IHS to core contact issue or if you didn't do a very good job applying the paste, then you would expect when the load went from 0% to 100% that the temperature for core 0 would really jump up but that's not the case. It moves far less than core 1 so that's why I think the problem is mostly sensor slope error related.

If you graphed these two curves on a piece of paper and extended the lines out, they would cross when the Distance to TJMax is at about 21 which is a temperature of about 79C. After this point, I'm not sure what will happen. Sometimes when there are two different curves, you'll reach a point where the two line up and then from 80 to 100 the two sensors might track each other almost exactly. I don't have enough experience with situations where one sensor is vastly different than the other to make an accurate prediction of what's going to happen higher up.

Your Core 1 sensor is the one to trust. It looks close to normal while Core 0 is definitely screwed up. rge did some thorough testing and came up with these guide lines:

http://www.xtremesystems.org/forums/...postcount=2429

For water cooling, RealTemp should report idle temperatures about 6C above your water temperature when you are running at 6x333 and 1.10 volts.
It looks like you are going to need a calibration factor of about +2.0 on core 1 to get it reporting correctly. For core 0, you are going to need a huge amount of negative calibration to get it to report some sensible numbers. You will need a difference of about 20 between the two so if core 1 is set to +2.0 then I would use something like -18.0 on core 0. Once core 1 is calibrated to a fixed temperature number then adjust core 0 so the two are balanced.

I think there is just barely enough adjustment room in RealTemp to take care of this problem. Now you can run some Prime 95 Small FFTs and see how they compare at full load. They should be a lot closer together from idle to full load which is how normal sensors respond on a 45nm Core 2 chip.

You could also bolt on a crappy air cooler and run some Prime Small FFTs with your fan turned off and take your CPU beyond 80C and see what happens to these sensors but in my opinion, that's kind of pointless. That would help answer the question whether TJMax is balanced on these cores at some point near TJMax but since you're water cooled, you're never going to be hitting these kind of temps with even a well overclocked and over volted Dual Core so there's not much point of that test.

Time to start overclocking that puppy. Another cool down test at about 4 GHz might tell me some more about what these sensors are doing higher up. When overclocking an E8400, 4GHz is always a nice place to start.

[Snafu]

Thanks Unclewebb. I'll give it a go and calibrate to core 1 with 25C idle temp being realistic with water cooling and RT @ 18C (36C load isn't too bad at stock speed under water).

Any chance Real Temp version 3.0 will include a slop adjustment? I am sure this would require far more testing per core to see what happens as temps get closer to TJmax.

One question, with the temps for core 0 being inaccurate, I am wondering whether the cpu is relying on this temp for throttling or whether there is another temp/sensor that it relies upon. I raise the point because even with recalibrating the sensor for core 0 to a more normal result, would the cpu throttle if the sensor read a high temp (albeit inaccurate)?

Will be calibrating and pushing the core. We'll see what happens.

Thanks again!

[unclewebb]

Thermal throttling is only based on what these two sensors are saying which probably doesn't give you a very good feeling right now.

Intel calibrates these sensors to be reasonably accurate for this purpose which is why two sensors operating on two different curves tend to line up somewhere around TJMax give or take a few degrees. I drew your temperature lines further and figured out that if the two stay on the same slope, then core 1 will reach the throttling point first so you shouldn't have to worry that core 0 is going to hold you back.

Temperature slope adjustment has been included in RealTemp since day 1. That's what entering in calibration factors does. It changes the slopes of these curves so they can better line up with each other. Your processor provides a good example of why this is sometimes needed. I think you're going to be surprised how normal your temps look after some calibration.

Will they be perfect? Probably not but they'll be a step in the right direction. With some more data at higher temperatures I might be able to fine tune your calibration a little further. Core 1 looks like a very good sensor with a wide range of motion. After checking your calibration again at the low MHz / low voltage point and following rge's guide lines, you won't have to worry about trusting this sensor. It will be very accurate from idle to TJMax.

On a 45nm Core 2 processor, you can run Prime95 on one core and nothing on the other and the idle core is still going to heat up to within a degree or two of the core that's doing all the work. The cores and sensors are so close together that the heat transfers to the other core very quickly even when one of them is idle. That's why as long as you have one core you can trust, you don't have to worry about the other one but after calibration, core 0 should track core 1 pretty closely.

[Snafu]

Good to know and I am glad that core 0 will likely not hold me back.

Once you enter a calibration factor would you adjust TJmax? I am noticing that with a calibration factor the sum of CPU temp and Distance to TJmax is no longer 95 (using beta v2.90). It appears to be out by the same value as the calibration factor.

By example (if needed), if a calibration factor of +2.0 is used then the sum is 97.

BTW the max values for calibration are +9.9 and -19.9.

[unclewebb]

The calibration factor makes it appear that the TJMax is changing from idle to TJMax. The Distance to TJMax data is the raw data coming from the sensors so I can't change that.

The basic formula is this:

Reported Temperature = TJMax - ( Distance to TJMax )

Since TJMax is a fixed number and the other value is raw data that I don't change, I added an extra part onto this formula. That's the calibration factor part.

If it appears that your TJMax has just dropped 20C at idle, that's because the calibration factor is giving you 20 degrees of correction at idle. As the temperature increases, the amount of correction the calibration contributes will decrease. A negative calibration factor will make it look like TJMax is less than 100C and a positive calibration factor will do the opposite and make it look like RealTemp is using a TJMax greater than 100C. If everything goes right, the amount of calibration will continue to decrease towards zero as these two curves finally meet.

[Snafu]

Okay sounds good.

Here are the cool down test results after calibration ( +2.0 for core 1 and -18.0 for core 0)



The results show that the lines are more parallel with each other. There still is a gap of ~25C but the lines are more parallel (graphed in excel).

Here is an idle shot (FYI) with both temps at 23C.



Now to start playing around.

[-X-hellfire]

I upgraded to the latest; RealTemp 2.90 beta, Rivatuner 2.22, Everest 4.60,
,,,but the OSD wont work in Crysis any more, it worked before with Rivatuner 2.11



When I upgraded RealTemp the calibrationfactors showed up as 0 0 0 0 which puzzled me somewhat as I had manually added them from the previous ini-file.

I noticed there were quite abit that had changed in layout of Realtemp, the APIC ID that showed core order is now gone, I´m just wondering if the calibrationfactors was set in the same way as before as the temperature looks different now, or is it just me


Will it help if I do a Prime 95 and post a screenshot of that?

[unclewebb]

X-hellfire: I hope all your problems are just you.

When you upgraded RivaTuner it looks like you put it in a different directory than what you were using before. You'll need to open up the Settings window in RealTemp, click on the RivaTuner button then click on Cancel to get rid of the previous link. Click on RivaTuner a second time and tell RealTemp where your new RivaTuner.exe is located now. After you exit the Settings window by clicking on OK, RealTemp will save your calibration factors and TJMax values into the correct RivaTuner Monitoring directory.

I'm using RivaTuner 2.22 and RealTemp 2.90 and the temps are showing up correctly on my G15. If they show up there but not on screen during a game then that sounds like a RivaTuner setup problem. Is it just Crysis that you are having this problem with or is it all games?

I have a Q6600 G0 like yours and I think TJMax = 100C is the best place to start.

Snafu: The cool down test is only showing the uncorrected data from your sensors so any calibrations you make won't change what that data looks like. Calibration is designed to make your reported temps look a little closer to reality. You can run a log file at a 1 second interval without the Excel option. Run a minute of idle, followed by a couple of minutes of Prime 95 Small FFTs, followed by a minute of idle. You should see your reported core temperatures of both cores move at a much more similar rate.

You can post that data here by using the code html tags surrounded by square brackets [].

use [c o d e] without the spaces, copy and paste in some data and then follow it by [/c o d e] again without the spaces.

[-X-hellfire]

X-hellfire: I hope all your problems are just you.

When you upgraded RivaTuner it looks like you put it in a different directory than what you were using before. You'll need to open up the Settings window in RealTemp, click on the RivaTuner button then click on Cancel to get rid of the previous link. Click on RivaTuner a second time and tell RealTemp where your new RivaTuner.exe is located now. After you exit the Settings window by clicking on OK, RealTemp will save your calibration factors and TJMax values into the correct RivaTuner Monitoring directory.

I'm using RivaTuner 2.22 and RealTemp 2.90 and the temps are showing up correctly on my G15. If they show up there but not on screen during a game then that sounds like a RivaTuner setup problem. Is it just Crysis that you are having this problem with or is it all games?

I have a Q6600 G0 like yours and I think TJMax = 100C is the best place to start.

Me too, but it's probably more people that upgraded Rivatuner etc, maybe I did something not so good, think I started Rivatuner and thought it would upgrade itself, but I aborted the installation when it said it wanted to make a Rivatuner 2.22 folder. Uninstalled 2.11 and restarted 2.11 installation but when it was started it froze, maybe some leftovers were left in the registry?
...anyway did a fix with Registry Mechanic, restarted and it worked like a charm, except the OSD thing

Upgraded Everest, but didnt uninstall it, it was installed in another directory, there was alot of textcolors etc that I didnt know how to save or migrate into the new one, might been the culprit to this problem, I really hate selecting textcolor one by one over and over again


I had already done the Rivatuner plugin selection in RealTemp and it was pointing to the new 2.22,

Tried Crysis and GRID with Rivatuner 2.11, but it only worked in Crysis, and then none of them


Okej, will change TJMax to 100 again, all the previous versions 2.69.5 had was set to 95, I might have missed something in this thread, tl;dr

Should I do a recalibration?

[Trigunflame]

@UncleWebb

Do these temperature variations look correct from core to core? If not could you suggest a tjmax/idle calibration based on this?

[unclewebb]

I originally used TJMax = 95C for the Q6600 G0 based on my IHS surface temperature measurement. With the help of rge's testing, I agree that number is likely too low and in order for the IHS to be at 95C, the actual core temperature has to be higher and 5C higher seems reasonable to me. That's where I originally went wrong and that's why I bumped it up to 100C.

For the G0, Intel lists the TJ Target as 90C. For my G0, that number doesn't seem to have anything to do with the actual TJMax. On core 0 and core 1, TJMax = 100C is reasonable but on core 2 and core 3, I think the actual TJMax is closer to 105C.

Intel's presentations last year confirmed that TJMax is not a fixed number and there is some variation from one CPU to the next of the same model. The 45nm Quads make it obvious that TJMax isn't fixed from one core to the next on the same CPU. I've seen variations of 10C from one core to the next which can't be explained by "slope error." If you take the time to post a Cool Down Test, I'll have a look at it and let you know what I see. If you've changed TJMax then you should check your calibration based on what rge found in this post:

http://www.xtremesystems.org/forums/...postcount=2429

With slope error and TJMax not being a fixed number, checking one's calibration is recommended. Even if the TJMax you use is not exact, calibrating RealTemp based on rge's guidelines will help minimize the amount of error in your normal temperature range.

When you abort an installation, it's hard to say what file and what version went where. Sometimes, depending on the installer used, System Restore can save your butt in a situation like this. The registry is the weakest link in Windows and is the reason why I avoid storing anything there. I like being able to delete or move folders without my registry getting confused.

Trigunflame: I don't own a Core i7 so I haven't played around with one yet. When I look at your results, core 0 and core 1 have almost a constant difference of 5C from top to bottom. My opinion is that actual TJMax for core 1 is likely 5C higher than core 0. If this was a slope error issue, this difference would narrow as you got closer to TJMax but in this case, it stays very consistent.

With Core i7, slope error seems to be a lot less of a problem but I think TJMax can still vary by up to 5C from core to core. Most of the screen shots I've seen, it's usually core 3 that reports the lowest temperatures, with core 0 the highest but yours is a little different.

For your CPU I'd just do a simple adjustment and set TJMax to 100, 105, 100, 101. After that you could try running at low MHz and core voltage and see how your reported core temperatures compare to your air or water temperature as mentioned in rge's post.

[Trigunflame]

Trigunflame: I don't own a Core i7 so I haven't played around with one yet. When I look at your results, core 0 and core 1 have almost a constant difference of 5C from top to bottom. My opinion is that actual TJMax for core 1 is likely 5C higher than core 0. If this was a slope error issue, this difference would narrow as you got closer to TJMax but in this case, it stays very consistent.

With Core i7, slope error seems to be a lot less of a problem but I think TJMax can still vary by up to 5C from core to core. Most of the screen shots I've seen, it's usually core 3 that reports the lowest temperatures, with core 0 the highest but yours is a little different.

For your CPU I'd just do a simple adjustment and set TJMax to 100, 105, 100, 101. After that you could try running at low MHz and core voltage and see how your reported core temperatures compare to your air or water temperature as mentioned in rge's post.

It had just seemed odd to me that there would be a 5c difference between cores; I didn't know if it was a problem on my end as far as seating and what not; being that I had lapped the IHS/HSF prior. Thanks for the enlightenment and program.

I appreciate the suggestion, I'll be sure to try those changes and frequency/vcore modification and monitor the results.

[rge]

Finally was motivated to measure core i7 IHS with calibrated thermocouple versus realtemp. Showed the same ~5C gradient from IHS to core as on E8400 and old pentium, using known tmax on core i7...though really was not surprised. Tried it at 4 different temps, but as usual only a few of 20 pics I took turned out with legible lighting for both computer screen and FLUKE thermocouple. At 70C RT, IHS was 65C. With heatsink on (hole drilled through heatsink with thermocouple on IHS), did not get temps higher than that...but was looking pretty linear.

IR was a no go...a heatsink is mandatory, these thing are up to 100+C even at .9vcore and 12x133, with just booting.

[Snafu]

IR was a no go...a heatsink is mandatory, these thing are up to 100+C even at .9vcore and 12x133, with just booting.

Holy Carp!

With heatsink on (hole drilled through heatsink with thermocouple on IHS

Do you have a close up shot of the base. I would like to give this a try. Thanks.

Apologies for the OT

[-X-hellfire]

I originally used TJMax = 95C for the Q6600 G0 based on my IHS surface temperature measurement. With the help of rge's testing, I agree that number is likely too low and in order for the IHS to be at 95C, the actual core temperature has to be higher and 5C higher seems reasonable to me. That's where I originally went wrong and that's why I bumped it up to 100C.

For the G0, Intel lists the TJ Target as 90C. For my G0, that number doesn't seem to have anything to do with the actual TJMax. On core 0 and core 1, TJMax = 100C is reasonable but on core 2 and core 3, I think the actual TJMax is closer to 105C.

Intel's presentations last year confirmed that TJMax is not a fixed number and there is some variation from one CPU to the next of the same model. The 45nm Quads make it obvious that TJMax isn't fixed from one core to the next on the same CPU. I've seen variations of 10C from one core to the next which can't be explained by "slope error." If you take the time to post a Cool Down Test, I'll have a look at it and let you know what I see. If you've changed TJMax then you should check your calibration based on what rge found in this post:

http://www.xtremesystems.org/forums/...postcount=2429

With slope error and TJMax not being a fixed number, checking one's calibration is recommended. Even if the TJMax you use is not exact, calibrating RealTemp based on rge's guidelines will help minimize the amount of error in your normal temperature range.

When you abort an installation, it's hard to say what file and what version went where. Sometimes, depending on the installer used, System Restore can save your butt in a situation like this. The registry is the weakest link in Windows and is the reason why I avoid storing anything there. I like being able to delete or move folders without my registry getting confused.

Thank you for clarifying this unclewebb


I have now done two Prime95 test, first one with many apps running then I thought maybe they influence the results and it would be interesting to see the difference with as few apps as possible, basically firewall & antivirus

I set the TJ Max to 100 and calibration factors to 0 0 0 0

[rge]

Holy Carp!


Do you have a close up shot of the base. I would like to give this a try. Thanks.

Apologies for the OT

Just put a power drill right in middle of the stock intel heatsink....drillbit will naturally stay in the middle as it is like a cup. Takes a couple minutes to get through the copper.

This pic i posted earlier in this thread when testing E8400..but couldnt find the post...but here it is.

Thermocouple needs to be calibrated. And need to use thermal paste like mx2, whatever to coat tip thermocouple to make good thermal contact with IHS.