Stuck CPU Sensor?

[GreatestOne]

I ran Real Temp 2.0 test Sensor and i gave a 0 and 7 for my 2 cores. So they said my Core 0 is "stuck." So...... what are you supposed to do about that, or what is it even? Didnt know CPU's had mocing parts on them! ...can i simply assume that both cores will be around the same temp so I can ignore this error?

 

[CamaroZ28]

Stuck sensor just means it won't be accurate for measuring temp unless its fully loaded. It's because the sensor in Intel chips is not designed to measure temperature of the core, but instead designed to shut the core off when it reaches a thermal limit. This is not a big deal as idle temperatures are not important when overclocking anyways because you can just watch the distance to TJmax on real temp to know if your proc is up to high.

 

[Frobot]

Well what are your temps for idle/load? If your temps even out under load, it's probably a stuck sensor, but if they both go up the same amount, my money would be on a bad HSF mount. I don't know how much stock initially I would put in that Real Temp sensor test...it gives me wildly different results every time I run it, and I have one core that's stuck for sure (+14 delta at idle!).

 

[CamaroZ28]

Well what are your temps for idle/load? If your temps even out under load, it's probably a stuck sensor, but if they both go up the same amount, my money would be on a bad HSF mount. I don't know how much stock initially I would put in that Real Temp sensor test...it gives me wildly different results every time I run it, and I have one core that's stuck for sure (+14 delta at idle!).

All the sensor test does is load the cores for a quick amount of time and then measure how much the temperature changes, and going off some values he got with a known good sensor compares those. Because the temperature of the cores is not going to be the exact same every time when it's loaded you are bound to always get different results in this test, so that is not abnormal.

 

[GreatestOne]

Well what are your temps for idle/load? If your temps even out under load, it's probably a stuck sensor, but if they both go up the same amount, my money would be on a bad HSF mount. I don't know how much stock initially I would put in that Real Temp sensor test...it gives me wildly different results every time I run it, and I have one core that's stuck for sure (+14 delta at idle!).

Idle: 35C, 29C (same as min temp)
TJ Max: 60, 66

During little activity the Core1 does fluctuate quite a bit and Core0 does seem frozen on the number until I put some more load on it. Then it lags behind jus a bit.

 

[jimbo6846]

Sounds like a stuck sensor. I'm not sure what that means, though. Like for example, will that mean that core might not shut down correctly if it gets too hot? I doubt it, but then again you never know.

 

[CamaroZ28]

Sounds like a stuck sensor. I'm not sure what that means, though. Like for example, will that mean that core might not shut down correctly if it gets too hot? I doubt it, but then again you never know.

No, this is common on these procs since these sensors are not designed to measure idle temps. A frozen sensor only refers to the sensor being frozen during low temps. These sensors are only designed to shutdown at a certain point, and this causes them to be less accurate at lower temperatures. Now if you were to fully load your core for hours and the numbers never changed at all, then that would indicate a real problem with the sensor, but since OP said that after load it starts to move some more, that is ok.

 

[unclewebb]

I don't know how much stock initially I would put in that Real Temp sensor test...it gives me wildly different results every time I run it, ...

Did you try reading the directions before running this test. Here's what the pop-up says:

For best results, makes sure your computer is idle before starting this test.

People that do scientific testing by ignoring the instructions will get results all over the place but if you follow the instructions, the results are very consistent from one run to the next. Your results will vary if one day you are underclocking your CPU and then the next day you have the voltage and MHz cranked up. If your testing environment is consistent then your results will also be consistent.

CamaroZ28 explained what the test is actually doing in his post above. It will help you discover when you have sensors that get stuck at lower temperatures. I've never heard of a sensor that didn't get unstuck and start functioning properly at higher temperatures. These sensors are designed for thermal throttling and thermal shut down control so I'm assuming that if these sensors were totally stuck, Intel would toss the CPU into the recycle bin.

 

[Frobot]

Did you try reading the directions before running this test. Here's what the pop-up says:

For best results, makes sure your computer is idle before starting this test.

People that do scientific testing by ignoring the instructions will get results all over the place but if you follow the instructions, the results are very consistent from one run to the next. Your results will vary if one day you are underclocking your CPU and then the next day you have the voltage and MHz cranked up. If your testing environment is consistent then your results will also be consistent.

CamaroZ28 explained what the test is actually doing in his post above. It will help you discover when you have sensors that get stuck at lower temperatures. I've never heard of a sensor that didn't get unstuck and start functioning properly at higher temperatures. These sensors are designed for thermal throttling and thermal shut down control so I'm assuming that if these sensors were totally stuck, Intel would toss the CPU into the recycle bin.

Yes, my computer is at idle. Yes, I know how to read. Yes, I know not to change 12 variables and wonder why my results are different.

I literally get a different result every single time I run "test sensor" at idle at stock settings. I can get 1-7-7-0 or 5-6-6-4 or 2-0-7-1. That's with no extraneous variables.

 

[CamaroZ28]

Yes, my computer is at idle. Yes, I know how to read. Yes, I know not to change 12 variables and wonder why my results are different.

I literally get a different result every single time I run "test sensor" at idle at stock settings. I can get 1-7-7-0 or 5-6-6-4 or 2-0-7-1. That's with no extraneous variables.

Try going into your bios and way underclocking your cpu and putting the voltage as low as possible to and see what happens.

 

[Frobot]

Try going into your bios and way underclocking your cpu and putting the voltage as low as possible to and see what happens.

I'll try that. But will I get different results considering I already have SpeedStep enabled? SS drops me to like 1.17v and 6.0x multiplier when I'm idling...

 

[jimbo6846]

I'll try that. But will I get different results considering I already have SpeedStep enabled? SS drops me to like 1.17v and 6.0x multiplier when I'm idling...

It wouldn't cause differing results every time you ran the test.

 

[Frobot]

It wouldn't cause differing results every time you ran the test.

Right, I mean would I get something different than manually going into the bios and selecting 1.17v and a 6x multiplier.

 

[jimbo6846]

If we're still talking about the Real Temp sensor test, you're probably not going to get the exact same numbers each time. The actual numbers you get aren't as important as the differences between the numbers in the result. For example, if you were to run three tests and get 6-7-8-8, 9-10-10-10, 8-6-7-9... that's perfectly normal. There isn't more than a 3-digit difference between the numbers in each set. If you get a bigger difference than 3, you probably have a stuck sensor. Also, I believe the test says if any of the numbers are below 2, that indicates a stuck sensor as well.

 

[Frobot]

If we're still talking about the Real Temp sensor test, you're probably not going to get the exact same numbers each time. The actual numbers you get aren't as important as the differences between the numbers in the result. For example, if you were to run three tests and get 6-7-8-8, 9-10-10-10, 8-6-7-9... that's perfectly normal. There isn't more than a 3-digit difference between the numbers in each set. If you get a bigger difference than 3, you probably have a stuck sensor. Also, I believe the test says if any of the numbers are below 2, that indicates a stuck sensor as well.

I'm not expecting the same numbers every time. But the thing has practically been a random number generator, with each core spitting out something between 0-7 every single time. Haven't gotten anything over 7 though.

 

[unclewebb]

Frobot: I apologize for doubting your reading skills.

Part of the problem might be that your computer isn't truly idle when you're testing. Some computers, especially ones running Vista with SuperFetch enabled, can have a lot of background tasks running that could make the RealTemp Sensor Test feature useless for you unless you find a way to turn them off. If you can post a screen shot of Task Manager with the graph that shows CPU Usage it might show us what's going on and why this test isn't giving you consistent results. You could also try turning on the RealTemp logging feature and set it to an interval of 1 second. After your computer has been sitting idle for a couple of minutes, stop RealTemp and have a look at the log file. When my computer is idle and the cores have reached their lowest temperature, there generally isn't any temperature movement.

To get consistent results out of this test, you need steady, consistent core temperature readings before clicking on the Test Sensor button.

 

[jimbo6846]

I'm not expecting the same numbers every time. But the thing has practically been a random number generator, with each core spitting out something between 0-7 every single time. Haven't gotten anything over 7 though.

Are you using the latest version? I use 2.70, the sensor test result variations are consistent for me.

 

[Frobot]

Frobot: I apologize for doubting your reading skills.

Part of the problem might be that your computer isn't truly idle when you're testing. Some computers, especially ones running Vista with SuperFetch enabled, can have a lot of background tasks running that could make the RealTemp Sensor Test feature useless for you unless you find a way to turn them off. If you can post a screen shot of Task Manager with the graph that shows CPU Usage it might show us what's going on and why this test isn't giving you consistent results. You could also try turning on the RealTemp logging feature and set it to an interval of 1 second. After your computer has been sitting idle for a couple of minutes, stop RealTemp and have a look at the log file. When my computer is idle and the cores have reached their lowest temperature, there generally isn't any temperature movement.

To get consistent results out of this test, you need steady, consistent core temperature readings before clicking on the Test Sensor button.

Oh, no worries unclewebb. I wasn't as upset as my reply sounded. My apologies. Thanks for your reply.

I think you're right that invisible background processes were affecting my results. After a lot of testing, the best way for me to obtain consistent results turned out to be opening CPU-Z, waiting until the multiplier went down to 6x and clock speed dropped to 2000 MHz THEN running the test at that moment. I did that about 30 times and then averaged the values for each of the cores, and I got pretty good results that way:

Core 0: 6.13
Core 1: 7.6
Core 2: 6.47
Core 3: 3.06

So I guess that one sensor is wonky? Here are my temps (Idle/Load, down at stock everything):

Core 0: 36c/55c
Core 1: 30c/51c
Core 2: 32c/53c
Core 3: 45c/59c

Can I pick your brain for a second? What do you make of that? I've already reseated the HSF once and the temp differences are still there. I want to say that Core 3 just has a bad sensor, but a +8 delta at load is still kind of bad, isn't it? My understanding is that under load they should more or less equalize...

 

[jimbo6846]

I think you're right that invisible background processes were affecting my results. After a lot of testing, the best way for me to obtain consistent results turned out to be opening CPU-Z, waiting until the multiplier went down to 6x and clock speed dropped to 2000 MHz THEN running the test at that moment. I did that about 30 times and then averaged the values for each of the cores, and I got pretty good results that way:

Core 0: 6.13
Core 1: 7.6
Core 2: 6.47
Core 3: 3.06

So I guess that one sensor is wonky? Here are my temps (Idle/Load, down at stock everything):

Core 0: 36c/55c
Core 1: 30c/51c
Core 2: 32c/53c
Core 3: 45c/59c

Can I pick your brain for a second? What do you make of that? I've already reseated the HSF once and the temp differences are still there. I want to say that Core 3 just has a bad sensor, but a +8 delta at load is still kind of bad, isn't it? My understanding is that under load they should more or less equalize...

While I can't exactly say that is "normal", it's definitely not an unexpected variation. Your temps actually starting to equalize out. The hotter the cores get, the closer the temps will get to each other. If you were to run Intel Burn Test, you'd see the temps get even closer to each other.

I have a feeling you'll get the same type of variation no matter how many times you reseat. That core probably isn't making the best contact with the IHS. It's unfortunate, but that's probably the situation.

 

[unclewebb]

Here's my E8400 running Vista when idle:

I think in XP it is even slightly better than that but even in Vista, the core temperature should be pretty steady when idle.

I'm not able to post links to the latest betas of RealTemp but if you go into Google and type in: RealTemp beta

a link to the latest version sitting on the FileDen site should pop up. The second item after the Google search at the moment.

The way Intel explained it at the last Developer's Forum in August, 45nm sensors have two main problems. The first problem is that these sensors can become "saturated" at lower temperatures so the sensor will simply stop moving even as the actual temperature of the CPU continues to decrease. Most users refer to this as a stuck sensor. It gets stuck as the temperature goes lower but when your CPU heats up and the temperature gets above the sticking point then it will function normally.

The second major problem is "slope error." What this means is that the raw data coming from the on chip digital thermal sensors changes at a different rate than what the actual temperature of the chip is changing at. When a sensor moves 40 positions, the actual core temperature of the chip may have changed by 30 degrees or by 50 degrees or maybe you are one of the very few lucky ones that has a CPU whose core temperature changed by exactly 40 degrees like we would all like.

I know plus or minus 10 degrees of accuracy sounds pretty extreme but you need to keep in mind that these sensors were only designed by Intel to provide thermal throttling and thermal shut down control. Intel doesn't recommend trying to get any use out of them beyond that. I found the plus or minus 10 degree spec when reading the data sheet for the 45nm Intel Atom line of processors. The sensors they are using on the Dual and Quad core E8000 or Q9000 chips might be of a better quality than that but after testing and seeing a lot of user results, I think they are probably very similar if not exactly the same.

For your Quad, core3 appears to be getting stuck and the other 3 cores operate on slightly different temperature slopes. The theory about trying to calibrate these is that if you lower your CPU MHz down to about 6x333 ~ 2000MHz and set the core voltage to about 1.10 volts and let your computer sit idle, a 45nm Quad should report core temperatures that are about 6C or 7C above your room temperature. You need to have your case open and your CPU fan on high if possible.

Use calibration settings in RealTemp to try and give you equal core temperatures during this test on your 3 cores and ignore the sticking one at the moment. Before version 2.75 of RealTemp I was assuming that these sensors would start to become accurate at around 60C but according to Intel, there is some unknown amount of error at TjMax and that error continues from one end of the temperature range to the other. The latest beta versions of RealTemp take this new information into account for the design of the calibration settings. Let me know how things work out for you.

It's possible when testing at low volts / MHz that you might discover a second core that has sticking issues.

 

[unclewebb]

On a Quad, core0 / core1 are joined and core2/core3 are also joined.

When running Prime95 small FFTs, your CPU at full load should show identical temperatures for the first two cores as well as identical temperatures for the second pair of cores but the two sets of cores are usually at slightly different temperatures. A difference of 0C to 5C is pretty typical at full load.

 

[jimbo6846]

Nice posts unclewebb. I've noticed my temps for the two different dies don't equal out until I hit mid-to-up 60's, which I can only reach with Intel Burn Test.

In fact, now that I think about it, I think even when Prime95 was able to reach temps that high (middle of summer, higher ambient temps) the temps still didn't even out. I think they've only ever evened out when using IBT. I wonder if that's due to the cores just working that much harder...

 

[unclewebb]

I use Prime95 small FFTs for temperature testing. It gives nice, consistent results as the temperatures build up or fall off when you stop running it. IBT creates more heat but when it switches from one round to the next the temps momentarily drop significantly before heading up again.

With Prime95 small FFTs, the load on each core should be very consistent so I've found that the actual temperature of each core of a Dual Core will be very similar if not identical.

With the way Intel explained slope error, it's possible that two uncalibrated cores might not equal out until TjMax and maybe not even then. There is an unknown amount of error at both ends of the temperature graph, idle and at TjMax, but Intel hasn't released any real documentation to show exactly how much error there is. If you draw two straight lines on a piece of paper, they may only cross at one point. By the chart that Intel showed at IDF, the amount of error appears to be fairly linear from one end to the other so if you use a linear correction factor like RealTemp uses I think you should be able to get a little closer to the actual temperature as long as your sensors are moving freely and not getting stuck.