RealTemp General Discussion

[wmw]

help

Hello everyone.
Need some help interpreting my results.
Here's the screenshot:

Some more information:
CPU: i5-3570k @4.5GHz
MB: ASRock Z77 Pro3
Case temp: 28ºC

What settings should I change to calibrate the temps?

Thanks in advance,
WmW

 

[unclewebb]

The problem with Ivy Bridge is that Intel cheaped out when they mounted the IHS to the cores. Here's some background info.

http://www.xtremesystems.org/forums...IHS-Removals-CPU-temp-dropped-from-79C-to-71C

Don't waste your time trying to calibrate the sensors on these CPUs. The sensors are not 100% accurate but there is no way to figure out if you have sensor error, a poor IHS issues or maybe heatsink mounting problems.

Enjoy your CPU and if you want to run it faster or cooler then you need to start cutting with the razor blade and hope that you don't go too far or you might end up with an expensive paper weight.

 

[wmw]

Thanks for the answer uncle.

 

[id0l]

Just wanted to say thanks for RT, I have been using it since the Core 2 launched and still prefer it over any other monitoring tool for CPU and GPU core temp. If I could ask for any improvement for the next version I would ask for improved system tray support for Windows 7, because you just can't get the CPU core order to save properly in the tray...you can manually move them around but it their position always resets on reboot. Then again I may be stretching here because I may have read a long time ago that was in issue with Win 7 and not RT. Still hella annoying! Haha.

 

[unclewebb]

I agree that the random placement of tray icons is annoying but this was a new "feature" added by Microsoft for Windows 7 and there is nothing I can do about it. RealTemp used to keep the tray icons properly organized in Vista but there is no way to do this in Windows 7.

 

[GeneO]

Realtemp modulating CPU

Hi,

This has probably been asked before but I haven;t been able to find a way to search this thread.

I use realtemp to monitor temperatures when stressing my CPU. I was finding lots of throttling messages in the Windows 7 kernel-power thermal operation event log every time I did a stress test:

Processor 2 was throttled by an entity other than the kernel power manager.
IA32_CLOCK_MODULATION MSR = 0x2.
Elapsed time since last event logged = 0s.
Log interval = 1000 events.

After ruling everything else out, I suspected an application and specifically realtemp.
Sure enough if I stress tested without realtemp, I saw no thermal modulation events.

Now I am just running realtemp to monitor temperatures, and not running the Sensor test.

My question is why is realtemp modulating my CPU clock?

TIA,
Gene

 

[unclewebb]

How do you have RealTemp setup? Post a screen shot of the Settings window.

In that window is a Clock Modulation option. Did you check that? What is it set to?

What type of computer do you have? Many laptops use Clock Modulation throttling when they are being stress tested with Prime95 or Linx / Linpack.

 

[GeneO]

How do you have RealTemp setup? Post a screen shot of the Settings window.
In that window is a Clock Modulation option. Did you check that? What is it set to?

What type of computer do you have? Many laptops use Clock Modulation throttling when they are being stress tested with Prime95 or Linx / Linpack.

Thanks for the quick response.

It is an Asus P8Z68-V PRO/Gen3 with a 2500k. It does not throttle when I run IBT without Realtemp running, and does when I have realtemp running. It is definitely realtemp related.

I have attached the settings.

 

[unclewebb]

There have been over 700,000 downloads of RealTemp 3.70 from TechPowerUp alone. If this was truly an issue with RealTemp, there would be more reports of this problem.

The guys over on the XtremeSystems forum push their CPUs as hard if not harder than any other forum. They do crazy stuff like remove the IHS off of their Ivy Bridge CPUs to get the last ounce of performance.

http://www.xtremesystems.org/forums...IHS-Removals-CPU-temp-dropped-from-79C-to-71C

Check that thread and see how many people there are using RealTemp. If RealTemp was responsible for throttling everyone's CPUs, I think these guys would notice and if they noticed, I know I would definitely be hearing about it.

How high is your core temperature during stress testing? Can you post a RealTemp screen shot of that? Also check off the Log File option in RealTemp and set the logging interval to 1 second. When you are finished stress testing, exit RealTemp and copy and paste the log file data to www.pastebin.com and then post a link here.

If your core temperature is OK then it might be something like the voltage regulator on your motherboard that is overheating and telling your CPU to throttle. Have you also tried testing with Prime95 using the Small FFTs option?

If you are convinced that RealTemp is the cause of your CPU using clock modulation throttlling then don't use RealTemp. I think that if you do some more testing, you are going to discover that there is something else going on here.

Edit: RealTemp reports the percentage of clock modulation throttling in the Status Area at the bottom.

When clock modulation throttling is in progress, the Load meter at the top will no longer be able to reach 100%. It is one of the very few load meters available that takes into account any internal CPU throttling. The Task Manager completely ignores this type of throttling and will continue to show your CPU at 100% even when it is being internally throttled to a fraction of its rated speed.

 

[GeneO]

In fact it the throttling events are not occurring under load (I think). All I have to do is start up real-temp and I get the system events. So is it something in the start up sequence?

 

[GeneO]

btw, don't get m,e wrong. I think realtemp is a great product and this doesn't seem to interfere with anything, I am just trying to understand what is causing these kernel events.

 

[unclewebb]

You just reminded me of something. Have you been running the RealTemp - Sensor Test? It uses Clock Modulation throttling to vary the load on your CPU during that test. It's been so long since I wrote that code that I completely forgot about it.

Hopefully that is the only time that RealTemp is doing this. If you don't run that test, your CPU should be running at full speed. After you exit that test or exit RealTemp, your CPU should also be returned to full speed.

Let me know if this fixes the problem you were having.

 

[GeneO]

No I haven't run it.

It is no big deal. It doesn't seem to affect performance. I will post if I can get better info for you or if I find it is some kind of odd coincidence or interaction.

Thanks

 

[unclewebb]

I had a look at the RealTemp code to try to figure out what's going on.

Clock modulation can be set to different values for each thread of each core of the CPU. The first time you start RealTemp, it disables clock modulation throttling across all threads and this choice gets saved to the INI file for future use.

After that, each time you start RealTemp, it restores the previously saved clock modulation values. If you don't use clock modulation throttling, RealTemp will ensure that this is turned off. If you do decide to set a clock modulation value of say 75% in the Settings window, RealTemp will restore this value to all threads.

The original problem was that the Dell XPS 1645 was setting clock modulation throttling for all of the threads except for the first thread. It might have been just a coincidence but to me it appeared to be done deliberately to avoid being detected. Most monitoring software at the time was assuming that all threads were being set equally so software was only checking the first thread. Based on this, I decided to equalize clock modulation for all threads and to turn them either all on or all off when RealTemp starts or exits just so the CPU is in a consistent state. Intel recommends that clock modulation be set equally across the entire CPU so I decided to follow that advice.

I wrote this code almost 3 years ago which is why I forgot about what RealTemp really does when it comes to clock modulation. Sorry about that.

These clock modulation adjustments happen when you start RealTemp as well as when you exit RealTemp so that explains why they are showing up in the event viewer. It's not being done to throttle or slow your computer down. It's only being done to make sure that your CPU is running at its full rated speed on each thread as Intel intended.

It does not throttle when I run IBT without Realtemp running, and does when I have realtemp running.

Can you check this again. If any one of your threads is using clock modulation throttling, RealTemp will report that in the Thermal Status area as shown in the picture above. RealTemp monitors each thread to make sure that it will be able to detect any problems if a manufacturer decides to secretly throttle only part of the CPU.

When you are running IBT, does RealTemp display Clock Modulation XX.X%

If not, your CPU is running as intended and is not being slowed down with clock modulation throttling.

 

[unclewebb]

http://img138.imageshack.us/img138/6/monitortest.png

Here's a good example of how various monitoring applications report clock modulation throttling.

In the top left corner, the Task Manager shows all 4 cores are fully loaded and the CPU is running at 100%. Prime95 is running on all 4 cores so that makes sense but have a closer look at the Prime95 results.

Worker #1 that is running on the first core is already on test 12 while the other 3 cores are only on test 2. There is a significant problem with this CPU that the Task Manager is completely ignoring.

Let's see what CPU-Z shows. It reports that all 4 cores are running at 4000.0 MHz. That's true but it doesn't say anything about 3 of the 4 cores being significantly throttled. HWiNFO64 tells the same story with 4 cores running at 4000 MHz. Core Temp confirms this too and also shows that the Load is 100% across all 4 cores.

Core Temp has the capability of reporting clock modulation but it only seems to check the first core. The first core of this CPU is OK so the Modulation box it shows is grayed out which would lead you to believe that the CPU is running at full speed.

RealTemp does 2 things right. It reports that at least one core is running significantly slower because Clock Modulation throttling is at 12.5%. The RealTemp Load meter shows 35%. The Task Manager load meter might claim that this CPU is running at full speed but internally, RealTemp shows that it is being throttled to a fraction of its Intel rated speed. The Prime95 results so far confirm that.

ThrottleStop goes one step further and shows exactly what the problem is. The first core is spending 100.0% of the time in the C0 state which means it is working as intended. CMod which stands for Clock Modulation is reporting 100.0 for that core which also confirms that the first core is running as it should be and it is not being throttled. The other 3 cores tell a different story. All 3 of them show serious throttling with Clock Modulation set at 12.5% and the C0% for each thread also confirms that there is some severe throttling going on within this CPU.

Kind of a long story about why RealTemp checks for Clock Modulation throttling. This was a serious issue for laptops before users had access to tools that could tell them exactly what their CPU was really doing internally. At the time, many of the popular monitoring tools could not detect this type of throttling and it looks like not much has changed in the last few years.

 

[Mussels]

so in summary - realtemp shows when individual cores throttle, due to temperature variations between the cores?

 

[unclewebb]

Clock modulation throttling has mostly been a laptop problem. There are two main types of throttling; one controlled by the CPU and one controlled by the chipset.

Laptops can use either type of CPU throttling and it is usually triggered by the GPU core temperature or overall power consumption. The CPU temperature rarely triggers this. It can also be triggered almost randomly where a CPU can be throttled to a fraction of its rated speed for extended periods of time without any real reason at all.

ThrottleStop was designed mostly for laptops and detects both types of clock modulation and shows exactly what threads or cores this is happening to. RealTemp will show a reduced C0% Load number if either type of clock modulation throttling is being used but in the Thermal Status area, it only reports the CPU clock modulation.

 

[ArsenicAcid]

So, I've had the same system for quite a while. Had no reason to upgrade anything. Cleaned it out and re-applied arctic silver on things as it was due for a clean-up refresh. Valid specs are -
evga 780i
Q6600 G0
Tuniq Tower 120 Extreme
Antec 902
Air temp entering front of case when test was done 10 minutes ago - 24c/76f, room temp averages 75-78 depending on time of day.

I've normally kept the processor overclocked @ 3.0-3.2 depending on the average room temps of whatever room I happen to have my desktop setup in (I rent apartments/houses). And this week have decided to see what all I can push the limits on. Normally I've used core temp, cpuid, cpuz hwmonitor and that's it. But today I got to reading most of the threads/info/reads regarding real temp, and now I've got a couple questions.

First, in hwmonitor, wtf are the temps in red? What is the deal with the temps in blue? I watched the whole process... at no point did anything in the value column reach the temps in blue. I know I'm not stupid, but is hwmonitor just wigging out to wig out? I even fully removed it from my computer and added fresh installs just to make sure it wasn't me.

Second - It's clearly obvious that cores 3 and 4 (2 & 3) sensors are sticking. But are my temps even close to where they should be? According to your guide, based on my specs I should be averaging 31-33 idle temps, not 34-38. Maybe it's late (4am) and I just missed something but I don't have a clear understanding in regards to how I'm supposed to adjust the tjmax. I've seen from a lot of places that the tjmax for my model/step should be 90 or 95. Default in Real Temp is 100. For shits and giggles I put it to 95 and it shot temps to 28c. Not sure if that was right or wrong since I've seen that intel doesn't disclose tjmax from some sources, and some sources said they had discussions with intel and intel gave a tjmax.

Anywho, based on the specs I've listed, I ran the sensor test. What exactly can you tell me other than what I already know, and that is that two of the sensors are reading wrong, and hwmonitor is giving me crazy ass shit that I can't decipher. Also, test was done on all stock settings for a baseline.

 

[unclewebb]

You will have to ask the programmer of HWMonitor what his utility is showing. Between you and I, I wouldn't worry too much about those numbers in the red or blue boxes.

If you run CPU-Z I think it will tell you that your CPU is a G0 stepping. I used to own one of these and I did plenty of hands on testing with it. My best guess is that the default TJ Max for this CPU is 100C. My next best guess is that the actual TJ Max can vary slightly from core to core with 100C being the baseline and actual TJ Max being slightly higher on some of the cores.

With actual TJ Max being approximately 100C on core 0 and core 1, actual TJMax on core 2 might be closer to 104C and actual TJ Max on core 3 might be about 103C. That's just my opinion on doing a pile of testing and looking at a lot of screen shots and data from other users.

If you do some overclocking and increase the core voltage, the amount of these differences will be clearer during the sensor test.

The reason I think Intel offset TJ Max slightly was to avoid all 4 cores reaching the thermal throttling temperature at the exact same time. Cores can be throttled individually in these processors. During a high temperature event, if the CPU can throttle the first 1 or 2 cores, this will help control overall heat and allow core 2 and core 3 to continue running at full speed so a user won't notice a sudden change in performance. If the core temperature continues to increase then a few degrees later, the CPU can throttle the other 2 cores as well.

I might be totally out to lunch here with this theory but it really doesn't matter. These temperature sensors are not 100% accurate temperature monitoring devices. They never have been and probably never will be. The only purpose of these sensors is to control thermal throttling at approximately 100C and thermal shutdown at approximately 125C to 130C. As long as these sensors are good enough for that, they are good enough. Intel never, ever intended these sensors to be used for 100% accurate idle temperature reporting so keep that in mind.

I don't see any evidence of your sensors sticking at idle. If you could eliminate some of the background crap running on your system, you should be able to lower the Load/C0% which will allow your CPU to idle down some more and maybe reduce the reported core temperatures another degree or two. Your room temperature is significantly hotter than during any testing I did and I always tested with an open case.

Your core temperatures look fine to me so continue overclocking and continue enjoying your computer and don't worry about the core temperature unless your CPU is unstable because it is running too hot. That's all that is important.

 

[ArsenicAcid]

Yes, it is G0 stepping. Now here's another question. Everything I've been reading says that tjcase temps would be the most accurate. I'm assuming my motherboard doesn't have a sensor for that or I would be getting a tjcase temperature somewhere.

 

[ArsenicAcid]

Did some fiddling around, cooler today here in Florida. Did quite a bit more research and found that indeed the q6600 has a tjmax of 90 per intel from some documentation a few years back. AIDA64 even says the same thing in auto settings. Also discovered in HWmonitor that CPU is infact the tjcase and cpu1 is motherboard temp.

Stable in prime95 for 2 hours before I shut it down, max cpu temp was 51, max core temps shown in real temp and screen shot @ 3.6ghz -

 

[unclewebb]

The only thing that controls thermal throttling in the Intel CPUs is the data coming from the core temperature sensors. The data coming from the CPU sensor is useless information whether it is 100% accurate or not.

The temperatures reported from the CPU sensor has nothing to do with the TCase temperature specification that is in the Intel documentation. The only way to correctly measure the TCase temperature of a CPU is to cut a groove in the top of the heat spreader covering the CPU and then you need to mount a thermal sensor at the geometric center of the heat spreader. This temperature spec was only intended for system builders looking to design cases and heatsinks and was never intended for end users unless you are willing to hack up your CPU.

The TJMax data that Intel released for these CPUs was a farce. When they first released some ridiculous numbers, myself and a few other programmers complained and a month later they released some new and improved numbers that were still not accurate. If you read the fine print of the TJMax numbers you found, you will see that Intel called these numbers TJ Target and then explained that actual TJMax might be higher. This basically meant that actual TJMax could be just about anything.

RealTemp uses 90C for the original B3 stepping Q6600 and it uses 100C for the newer Q6600 G0 stepping like you have. The recommendations I gave you above will result in the most accurate core temperatures but you can use whatever TJMax you like.

Edit: Have a look at what AIDA 64 reports for your CPU temperature. It shows 37C so it makes perfect sense that the core temperature which is measured at the hottest spot on the core is going to be higher than 37C. If you use TJMax =100C, RealTemp will show 42C. That makes sense. A core temperature less than 37C doesn't make any sense.

Edit #2: If the multiplier reported in RealTemp is not accurate then exit CPU-Z and all your other monitoring programs. I think CPU-Z has started using some different system timers. This interferes with RealTemp's results since Intel uses shared monitoring timers in their CPUs that any monitoring software can use however they wish.

 

[Growltiger]

RealTemp has a bug with the Max in the tray

RealTemp doesn't work quite as it should any more. A bug?

I just installed the latest version, 3.70. I also set it to show the Max instead of Core 0. Just to complicate things further I also upgraded to Windows 8 Pro.

When in a window it is fine. But when it is minimised to the tray, the Max temperature stays there for a bit and then vanishes.

The Core 0 and GPU temperaure work correctly, they always stays there. But if you want the Max it just vanishes. If you right click on one of the others and click Redraw icons, it reappears for a bit and then vanishes again.

 

[Growltiger]

The bug - more info

The bug is a bit more complicated than I thought. I left it running showing three temps: GPU, Max and Core 0. When I came back there were only two showing, but this time the one that vanished was the GPU. Redraw Icons won't bring it back.

When restored to a window, everything is normal.

 

[unclewebb]

In Windows 7 you have to tell the OS to always show all icons for this RealTemp feature to work correctly. I haven't played with Windows 8 recently to see if anything has changed.