RealTemp General Discussion

[unclewebb]

RealTemp is a utility to monitor core temperatures on all Intel Core based processors.

To learn more about RealTemp and to download the latest official version then head here:

http://www.techpowerup.com/realtemp/

The current beta version is 3.69.1
I will update this as new versions become available, unless I forget for a few months.

The beta has Sandy Bridge support and can be downloaded here:

RealTemp 3.69.1
http://www.mediafire.com/?4uixpjtezznuzkd

Recent additions include Core i7 support and access to Clock Modulation that is built into Core processors.
Version 2.84 has added a new sensor test which does a more thorough job of checking for stuck sensors.
Version 2.86 has a bug fix for Core i7 when hyper-threading is turned off.
Version 2.90 RC2 has a new layout, including a Load meter.
Version 2.90 RC8 includes Nvidia GPU temps.

If you have any questions or comments about this program then here is the place to do it

Cheers.

 

[Laurijan]

Is this program more accurate then core temp i ask because i get differnt readings with this program?

 

[unclewebb]

Here's a link to some of the testing I've done.

http://www.xtremesystems.org/forums/showpost.php?p=3096431&postcount=1600

Intel does not publish TjMax for any of their Desktop processors. RealTemp has tried to come up with this value from actual testing with an IR thermometer. The competition does not seem to have done any real world testing. If they had, they wouldn't still be using TjMax=105C for the 45nm desktop processors.

 

[unclewebb]

Just a quick update so that you can drag RealTemp while in Mini-Mode.
Special thanks to WoZZeR999 for complaining about this oversight. :up:

You can download it here:
http://www.fileden.com/files/2008/3/3/1794507/RealTempBeta.zip

Unzip and copy the new RealTemp.exe into your present RealTemp folder. You will need to download version 2.60 first and install it if you aren't already using RealTemp.

You enter or exit Mini-Mode by double left clicking on the GUI.

Even the truly anal will have a hard time complaining about RealTemp taking up too much screen real estate now. Throw it in Mini-Mode and then drag it where ever you like.

RealTemp can also start up in Mini-Mode as well.

The other feature added to this version is that a double right mouse click on the GUI will move RealTemp up to the upper left corner of your screen so it's out of the way. I might add an option later so you can choose where on the screen you want it to go after a double right click.

Feedback is always appreciated.

 

[Kreij]

@unclewebb

The first line of the documentation reads...

Real Temp is a temperature monitoring program designed for all Intel single Core, Dual Core and Quad Core processors.

You may want to clarify that it's only the Core and Core2 CPUs as they are the ones with the thermal sensors in them.

Yes, I wasn't thinking and tried it on my Pentium D
At least it was nice enough to tell me it wasn't supported without being insulting.

 

[unclewebb]

By capitalizing Core I was implying that it only works on the Intel Core architecture of CPUs. That's definitely a little cryptic so I'll try to add "No Pentiums" to the docs in the near future. The Intel documentation is a little unclear. I've read that the very late P4 processors actually had DTS sensors in them which are needed for RealTemp to work. I'm still waiting to get some feedback on this since I was all AMD during the P4 era.

 

[unclewebb]

Users are sometimes a little unclear about how to go about calibrating RealTemp. It's not an exact science. I recommend lowering the MHz down to about 1600 MHz (266 MHz x 6.0) and the core voltage down to 1.10 volts. If you are using a 45nm processor with a default FSB of 333 MHz then use that instead. I also open the case to get rid of that variable. The goal is to lower the CPU heat output to a minimum and establish a common reference point to the testing I've done.

If you are unsure about manually setting MHz and core voltage then you can also enable C1E in the bios. If you are not overclocking, after your computer boots up you should end up with similar values to the above when Windows is idle. Use CPU-Z to confirm that your motherboard correctly supports C1E.

With water you'll have to compare the RealTemp reported core temperature to your water temperature so give it some time to stabilize. If you are air cooled then you need to compare your idle temperature to your air temperature upwind of your CPU.

Here's a quick test I did a while ago with my Q6600 - G0. I usually head downstairs to my cool basement but decided to give it a try upstairs where the temperature beside the open computer was 21C and in front of the computer it was closer to 20C. The side panel was off but it's sitting beside a desk so airflow isn't perfect. I'm using calibration factors of
(1, 1, 2, 1 )
which typically has this processor within +/- a degree or two at idle. Here are the results:

Average core temperature was 26C and room temperature is about 20C to 21C.

Since this test I've added one digit after the decimal point to the calibration factors so there are more options to choose from. At idle, during this test, you should see pretty much the same temps across all 4 cores.

 

[unclewebb]

RealTemp 2.69.3

Finally had a chance to add in an alarm feature to keep CompuTronix happy.

It's only a visual alarm at the moment where you will see the application and the RealTemp system tray icons flash. I will be adding an alarm that you can also hear as soon as I find an appropriate .wav file.

I've found that Quad core processors at full load will usually have the first two cores reporting higher temps than core2/core3 so I've created the ability to set two different alarm points if you need to compensate for that. Single and Dual core processors will only have one alarm temperature to choose. The temperature range is from 0C to 125C and if you don't want an alarm then just leave it unchecked. You won't likely ever hear an alarm if you set it to 125C since that is believed to be the thermal shut down temperature of the desktop processors.

It's available for download in the beta section:
http://www.fileden.com/files/2008/3/3/1794507/RealTempBeta.zip

Download, unzip and copy the new RealTemp.exe into your RealTemp directory.

To make room for this option in the Settings window I moved the Core Order to the main screen. It has been renamed APIC ID which is the technical term for it. If you have a Dual core you won't see this information since your APIC ID will always be 01. This information is only used for the grouping of the cores of a Quad core processor.

The other small feature added was a user selectable anchor position.

By default, if you do a double right mouse click on the RealTemp GUI, it will jump to the top left corner of the screen. If you would like it to jump to a different location then position RealTemp where you want it, hold down the Shift key and do a double right mouse click while holding it down. This will set a new anchor position.

Now when you double right click on RealTemp it will jump to this new custom anchor position that you have chosen. I realize that setting a new anchor position is a little cryptic but I'll try to document it early on in the New Features section. Other than that I guess it will be a special feature for users that actually RTM!

If anything isn't working quite right with these new features then let me know. If you have any sounds that you would like to hear for an alarm then e-mail them to me. I'll try to code the final version so users can custom choose their own .wav file to play for an alarm.

 

[stasio]

Link above in the beta section,still ver.2.69.2

 

[unclewebb]

Link above in the beta section,still ver.2.69.2

That seems to be a problem with the FileDen site I use and usually Firefox. You will sometimes have to go in and clear your cache within Firefox before doing a download so you get the latest version. The cache can be found under Tools -> Options -> Advanced in Firefox. I just did a download and the latest version is there.

 

[goren]

Hi,
Sorry if this has been answered somewhere. I've checked the documentation and haven't found the answer.

How exactly did you determine TJMAX? Did you point the IR thermometer at a bare CPU as it heated up until the thermal protection kicked in?

How accurate is measuring the top of the IHS for determining core temperature or TJMAX?

Also, is TJMAX identical for all Core processors (65nm or 45nm, dual or quad) and what about the newer Pentium Dual cores (E21xx and E22xx)?

Thanks!!!

 

[unclewebb]

goren: TjMax was determined by pointing an IR thermometer at a variety of Core based processors, 45nm and 65nm as well as Dual and Quad core. All measurements were taken while the processor was idle with as little load as possible.

When a CPU is running an application, hot spots develop at various parts within the core depending on what type of instructions are being executed. At idle, these hot spots are greatly reduced and Intel's testing has shown that the temperature measured at the center of a CPU similar to how I am measuring is typically within 0C to 1C of a core temperature reading. For this reason, I believe that a measurement of the IHS with an IR thermometer can get you a very accurate approximation of the core temperature.

Here are a couple of good papers from Intel about core temperatures:
http://download.intel.com/technology/itj/2006/volume10issue02/vol10_art03.pdf
http://arxiv.org/pdf/0709.1861v1

rge at XtremeSystems has been gathering info for a long time so if you're interested in learning more than I would try contacting him.

You don't have to go all the way to TjMax before you start to see a very strong relationship develop between the digital thermal sensor (DTS) data and the changes in the core temperature as measured with an IR thermometer. Typically by 60C and up there is a direct 1:1 relationship which is obvious when the IR thermometer is pointed at the IHS.

The basic formula is this:

Reported Temperature = TjMax - DTS

If we rearrange that we get:

TjMax = DTS + Reported Temperature

If the DTS is showing 35 while an IR thermometer is showing 60C then TjMax must be equal to 95C. The same thing happens when you get to 70C. Now the DTS is showing 25 so you add them together and TjMax must be 95C. As the temperature cools off and you get farther away from 60C, the 1:1 relationship starts to fall apart and you will see either the temperature start to change faster than the DTS is changing or vice versa where the DTS is changing at a different rate than the temperature is changing at. The long standing belief that the output from these sensors is 100% linear doesn't hold true at idle. This is explained further in the RealTemp documentation.

You don't have to take a processor all the way to the throttling point near DTS=0 to calculate what TjMax is but let's just say that more than one of my processors has experienced a very wide temperature range.

The easiest processor to test was the E2160. Most of the heat at idle seems to be created by the on die cache memory. This processor only has 1MB of L2 cache so it heats up very slowly even without a heatsink on it. The extra cores and extra cache on a Q6600 make testing more difficult but it also showed a TjMax=95C relationship during testing.

So far I have only found two possible TjMax values for the Desktop processors. It's either 85C or 95C for the Desktop chips. For the mobile processors it is either 100C for the 65nm chips or 105C for the new 45nm mobile chips.

Here is a link to the RealTemp forum at XS and some of the testing I've done:
http://www.xtremesystems.org/forums/showpost.php?p=3096431&postcount=1600

Here is some testing that rge at XtremeSystems did and his conclusion about TjMax and whether measuring the temperature of the IHS is valid at idle:
http://www.xtremesystems.org/forums/showpost.php?p=3085792&postcount=1525

 

[P4-630]

Everest Ultimate reports a 105C (221F) Tjmax Temperature for my E7200,
should I change the Tjmax from 95 to 105 to get the most accurate temps?

 

[unclewebb]

If you believe that TjMax=105C for your 45nm E7200 then you can use that in RealTemp. Personally, I don't believe that for a second. Everest also uses TjMax=105C for my E8400 and after this test, I don't believe that either.

http://www.xtremesystems.org/forums/showpost.php?p=2883315&postcount=573

Here are some comments from the programmer of Everest:
http://www.xtremesystems.org/forums/showpost.php?p=3161151&postcount=1785

And here is some more testing of a 45nm E8400:
http://www.xtremesystems.org/forums/showpost.php?p=3085792&postcount=1525

TjMax=105C is only documented for the Intel mobile processors. There is no documentation from Intel that says it is relevant for any of the desktop processors.

The default TjMax used by RealTemp is based on IR thermometer testing. Read the documentation here on TechPowerUp and try doing the calibration test. You will likely agree that it is impossible for your E7200 to have a TjMax=105C.

 

[burebista]

unclewebb first of all THANK YOU for you superb piece of work. I'm an E8400 owner about 3 days now and I read alot on the Web about Wolfdale temperatures. Finally I fully agree with your approach and methodology so I'm a happy RealTemp user.
Now I have two questions for you. In picture below is my setup: E8400 under Scythe Ninja fanless on ASUS P5Q-E MB. Two case fans, Noctua P12 exhaust at ~700RPM and Scythe Kama PWM intake at ~ 500 RPM.



At room temperature of ~23°C and E8400 at 0.944V Vcore and underclocked to 1586MHz (264x6) those are temperatures reported by RealTemp and from BIOS via SpeedFan (in SpeedFan I've calibrated core temperatures to match RealTemp readings).

IMO idle temperatures reported by RealTemp are too high, I tend to believe BIOS CPU temperature of 26°C.
I ran your sensor calibration test and give this



So for me it seems that my Wolfdale show a wrong idle temperatures on cores. I can't believe that an E8400 with 0.944V Vcore and 1586 MHz have 35°C in idle, it seems more accurate that 26°C from BIOS.
I guess my sensors are stuck on idle temperatures?
In full-load RealTemp looks OK.



55°C peak on 3GHz under Ninja fanless, summer, low RPM's fans it seems quite right to me. I touch my Ninja's heatpipes and they was warm. Not hot but pretty warm so I guesstimate that 55°C was real.
So my question is can I ignore idle temps (I've tried to calibrate them but after that your sensor test show 0 movement on cores) and consider only load temperatures (in fact that's what I need to know, load temperatures to be OK)?

And my second question is about differences in CPU frequency marked with red box. In CPU-Z is correct, but RealTemp (it's odd, only sometimes is wrong) show me another frequency. Do you have any ideas why?

Thanks again for your time.

 

[P4-630]

If you believe that TjMax=105C for your 45nm E7200 then you can use that in RealTemp. Personally, I don't believe that for a second. Everest also uses TjMax=105C for my E8400 and after this test, I don't believe that either.

Sorry but I did not mention anything about what I believe or not.
It was just a question, since there is an option to change the Tjmax in RealTemp.
But I understand for the most accurate temperatures you recommend to leave the Tjmax at 95 for the E7200.

 

[unclewebb]

burebista: What is reported as CPU: temperature in SpeedFan may be a believable looking number but without calibration, you can't know for sure how accurate that number really is so avoid making any comparisons to it. On my board, the CPU reading at times is off significantly.

The sensor test shows that your sensors are not stuck at this temperature which is good. These sensors are rarely accurate at reporting idle temperatures so if your sensors read a little high that is not unusual. My E8400 does the exact same thing.

If you had a high speed fan on your CPU cooler than I would recommend that at idle during your low MHz / low voltage test that your reported temperatures should be about 4C or 5C higher than your room temperature in an open case. If reported temps are higher than this then you just need to go into the Idle Calibration section of RealTemp and enter in a negative calibration factor. I think my E8400 needed a calibration factor of approximately -2.0 so the idle temps were more accurate.

This calibration procedure will improve the accuracy of your temperatures from idle to about 60C. These sensors are usually very accurate by this level so even without any calibration, your load temps are probably within a degree of the actual temperature.

When running a CPU cooler fanless, the idle temps will be higher than normal. It's really hard to know just how much higher without knowing about your overall case air flow and 101 other variables. Run a high speed fan on that cooler and it will be easier to compare to what I have measured during this test.

The CPU MHz in RealTemp is a work in progress. It does an excellent job on most motherboards most of the time but it's not perfect. Looks like CPU-Z is still king of MHz so if there is a significant difference, I would trust that program first. The RealTemp MHz were correct on your other screen shot. How often would you say is RealTemp wrong? After I get the next version finalized I can work with you and your motherboard to help improve the accuracy of RealTemp MHz if you like.

P4-630: I apologize for the misunderstanding. When RealTemp first came out I had to do a lot of defending of my methods but the initial disbelief has changed in the last few months. Some users like RealTemp but disagree with my choice of TjMax so the adjustable TjMax feature is for them. It's greyed out to discourage users from making any changes since I don't know of any CPUs at the moment that need a change of TjMax.

 

[burebista]

Thanks man for your explanations. I'll try to put a fan on my Ninja to see how it looks like in idle then.

The RealTemp MHz were correct on your other screen shot. How often would you say is RealTemp wrong? After I get the next version finalized I can work with you and your motherboard to help improve the accuracy of RealTemp MHz if you like.

I have ASUS's EPU software which have 4 profiles, from maximum power savings to turbo mode.
If I launch any of these profiles then I bring CPU in full-load (Orthos for example) RealTemp is fine, it shows like CPU-Z same frequency.



But if I use Maximum power saving profile from EPU after a while RealTemp go crazy.



If you have time to spend with me I'll be glad to help you. Maybe it will be useful for others too. EPU, DES are standard now (at least from Gigabyte and ASUS) and maybe somehow RT is fooled by that behavior.
I know that your main focus is temperature related, but when you find some free time I'm here to help.

 

[unclewebb]

I'll have to look into what exactly Asus EPU software is doing in Maximum power saving mode. Is it possible that RealTemp is measuring something internally that is going on during this mode that CPU-Z is ignoring? If I learn something new about this I'll post the results.

When not running the Asus software, is RealTemp correct at idle and at full load?

 

[burebista]

When not running the Asus software, is RealTemp correct at idle and at full load?

Damn I must uninstall it and see. I can't just exit from it.

Anyway, now it's fine.



Weird.

 

[burebista]

OK, I've stopped EPU from Task Scheduler, reboot and I make a short movie with what's happens on idle, on Orthos launch and again in idle.
Maybe it helps.

 

[unclewebb]

Your video definitely helps. At full load the MHz look pretty steady but at idle when SpeedStep / C1E is cutting in and out it's confusing RealTemp. I'll send you a slightly different version of RealTemp later this week to see if I can improve the accuracy for you at idle. Jumping back and forth between 1999 and 2999 MHz is normal when C1E is enabled but all those other numbers are not.

 

[P4-630]

unclewebb:
Fluke does not include a calibration certificate with these meters but these Fluke 62 Mini Infrared Thermometers are factory calibrated and measure with an ± 1.5 ºC accuracy so your test results should be pretty accurate.

http://us.fluke.com/usen/products/s...965(FlukeProducts)&category=THE(FlukeProducts)

 

[unclewebb]

At the idle temperatures I was measuring at, the accuracy is even a hair better than that. It's listed as ± 1 ºC in the Fluke docs.

10 ºC to 30 ºC (50 ºF to 86 ºF): ± 1 ºC (2 ºF)

 

[unclewebb]

And my second question is about differences in CPU frequency marked with red box. In CPU-Z is correct, but RealTemp (it's odd, only sometimes is wrong) show me another frequency. Do you have any ideas why?

I've been working with burebista the last couple of days to try and figure out what's going on. Thanks to the incredible amount of testing he did, I think we've come up with a reasonable explanation.

At full load, RealTemp was showing the exact same MHz as CPU-Z. At idle it would also show the exact same MHz but after a while RealTemp would start to report MHz that frequently changed and at times would drop to only about half as much as CPU-Z was reporting. He managed to track this down to having C-State Tech enabled in his bios along with either C1E or SpeedStep enabled.

C1E drops your FSB multiplier down to 6.0 when your CPU is idle but with his motherboard and processor, it seems that the it was going further and actually stopping and turning a core on and off depending on the load. These deeper sleep states were previously only available on the mobile chips but it looks like the newer desktop chips with the appropriate motherboard can also go into these states. (C1,C2,C3,C4, etc.) I use RMClock 2.30 to experiment with these. Very few desktop CPUs and boards support all of these modes.

CPU-Z seems to report the MHz based on when the processor is running. RealTemp averages when it is running with when it is sleeping so the MHz it reports when this is happening is lower than what CPU-Z reports.

If the processor is throttled down to 266.6 MHz x 6.0 then CPU-Z will report about 1600 MHz. During the interval that RealTemp uses to calculate MHz the processor might be at 1600 MHz for 75% of the time and it might be asleep at 0 MHz for the other 25% of the time. If this happens, RealTemp will calculate the average during this interval and report it as 1200 MHz.

I think both of these answers are half right. If RealTemp reports your idle FSB MHz lower than what you have set in the bios then it is very likely that a power saving mode has kicked in and is letting one or more of your cores have a snooze.

Thanks for all your testing burebista. We flipped a coin to try and decide whether this is a RealTemp bug or a feature. He agrees that in some ways it is a feature because it indirectly informs users when parts of their processor is going into a sleep state. I'm leaning towards letting RT continue to report an averaged MHz.

Here's an example of what he sees when his computer is very idle: