RealTemp General Discussion

[unclewebb]

JBravo: A lot of the 45nm sensors have problems reading correctly at low temperatures like 20C or 30C. Many of them get stuck.

If your room temp is 20C then try running a cool down test at default MHz and core voltage with Speedstep or C1E enabled so at idle the CPU drops down to a multiplier of 6.0. Post or send me your results so we can start by having a look to see if either of your sensors are getting stuck.

 

[JBravo]

JBravo: A lot of the 45nm sensors have problems reading correctly at low temperatures like 20C or 30C. Many of them get stuck.

If your room temp is 20C then try running a cool down test at default MHz and core voltage with Speedstep or C1E enabled so at idle the CPU drops down to a multiplier of 6.0. Post or send me your results so we can start by having a look to see if either of your sensors are getting stuck.

Thanks for the reply

I've went ahead and did the calibration that stated earlier but its not working properly I think.

I took the speed and multi down to 266x6 @ 1.1V,booted to windows. Then I set the minimum temp in realtemp down to the ambient which was 17C and added 12C to be safe giving me 29.

Now, I thought this was supposed to give me the "lowest" possible readings I would ever see, BUT when I set it back to 3.3ghz @ 1.225V, guess what, still got 29 idle...And yes, the one sensor is sticking badly,barely moves ( core 0 ),but core 1 moves instantly.

The core 0 sensor does however move when temps go up to about 45-50 and normally lags behind by 5C or so.

Any help would be appreciated very much Unclewebb!

 

[unclewebb]

Rule #1 is you can't calibrate a stuck sensor using my method. Stuck is stuck.

You still need to do a full CPU Cool Down Test and show me your results. Both core 0 and core 1 get stuck on my E8400 at slightly different levels so we need to check to see if core 1 is getting stuck also on your CPU. Your room temperature is cooler than most so there's a good chance that it's going to be stuck too but at a lower temperature. Not many of these sensors are capable of getting down to 25C or lower.

Head to www.imageshack.us and upload a screen shot of your test there and then post a link to it here.

 

[JBravo]

Rule #1 is you can't calibrate a stuck sensor using my method. Stuck is stuck.

You still need to do a full CPU Cool Down Test and show me your results. Both core 0 and core 1 get stuck on my E8400 at slightly different levels so we need to check to see if core 1 is getting stuck also on your CPU. Your room temperature is cooler than most so there's a good chance that it's going to be stuck too but at a lower temperature. Not many of these sensors are capable of getting down to 25C or lower.

Head to www.imageshack.us and upload a screen shot of your test there and then post a link to it here.

Here is the sensor test as requested Unclewebb. Sorry for taking so long to respond. I'm not registered at imageshack so I'll just post the here?

The sensor on core 1 seems to be responding to each change in temp, so I'm hoping thats the one I'm going to use. On the screenshot the minimum shows as 29, but I set the calibration back to 0 and didn't reset the temps afterward. The minimum should be 41 with no calibration, as shown again in the idle state. Hope this is what you wanted,and thanks for all the help thus far.

 

[unclewebb]

You don't need to be a member at Imageshack. You just upload images and they give you a link to them. I'm not sure how long they will keep images on their site if you're not a member but it seems to be for 6 months or a year or even longer.

See why I like the CPU Cool Down Test. It makes it very easy to spot a stuck sensor and your core 0 sensor is definitely stuck.

Looking at your numbers, I think somewhere around 70C or 75C, core 0 will catch up to core 1 and they will likely remain fairly equal from that point to TJMax. That's usually what happens.

No matter what you do, the idle temperature of core 0 is going to be a meaningless number. It's stuck at idle and there's nothing you can do about that except ignore it.

rge's testing was with an open case. If you also tested with an open case and your room temperature near your case is 17C then 29C or 30C sounds like a reasonable temperature number for your CPU. Most users don't use the OEM cooler for very long so I haven't done much testing but rge's numbers / guidelines seem OK to me.

It's typical for these sensors to read either too low or too high at idle but most of them are a lot more accurate at higher temperatures. A 10C degree error at idle is not unusual. Using a calibration factor somewhere around -8.0 for core 1 seems about right. Leave TJMax set to 100C for both cores. If your case was closed when you tested then you might need a little less calibration factor than that. The closer you duplicate rge's testing environment, the better your results will be.

After that I'd run Prime95 Small FFTs at your normal or overclocked speed and adjust the calibration factor on Core 0 to get the cores to line up reasonably well. It will probably need a positive calibration factor. Either that or just leave core 0 as is knowing that it's not a sensor you want to trust too much.

 

[JBravo]

You don't need to be a member at Imageshack. You just upload images and they give you a link to them. I'm not sure how long they will keep images on their site if you're not a member but it seems to be for 6 months or a year or even longer.

See why I like the CPU Cool Down Test. It makes it very easy to spot a stuck sensor and your core 0 sensor is definitely stuck.

Looking at your numbers, I think somewhere around 70C or 75C, core 0 will catch up to core 1 and they will likely remain fairly equal from that point to TJMax. That's usually what happens.

No matter what you do, the idle temperature of core 0 is going to be a meaningless number. It's stuck at idle and there's nothing you can do about that except ignore it.

rge's testing was with an open case. If you also tested with an open case and your room temperature near your case is 17C then 29C or 30C sounds like a reasonable temperature number for your CPU. Most users don't use the OEM cooler for very long so I haven't done much testing but rge's numbers / guidelines seem OK to me.

It's typical for these sensors to read either too low or too high at idle but most of them are a lot more accurate at higher temperatures. A 10C degree error at idle is not unusual. Using a calibration factor somewhere around -8.0 for core 1 seems about right. Leave TJMax set to 100C for both cores. If your case was closed when you tested then you might need a little less calibration factor than that. The closer you duplicate rge's testing environment, the better your results will be.

After that I'd run Prime95 Small FFTs at your normal or overclocked speed and adjust the calibration factor on Core 0 to get the cores to line up reasonably well. It will probably need a positive calibration factor. Either that or just leave core 0 as is knowing that it's not a sensor you want to trust too much.

Thanks for having a look UncleWebb

That is exactly what I was doing, and my calibration was set to -8.0
And yes, I'll just forget about core 0 and work with core 1.

Thanks again

 

[Randallizer]

It is probaly me but

I have the Intel i7 920 and when I run version 3.0 it only shows two cores. On the other hand if I run version 2.70 it shows the 4 cores. Is there something I'm not doing right?

Thanks in advance for your answer

Randallizer

 

[unclewebb]

RealTemp 2.70 does not read the Core i7 correctly but 3.00 generally does. Can you post a screen shot of RealTemp and CPU-Z?

The latest beta of RealTemp is located here:
http://www.fileden.com/files/2008/3/3/1794507/RealTempBeta.zip

Use that version to see if this bug has been fixed already.

One user has an issue where it is his operating system that limits the number of cores RealTemp can see. With some more info, I'll help you get RealTemp working properly. Let me know your OS and whether it is 32 or 64 bit.

 

[Randallizer]

Thanks for taking a look, following your comment about the motherboard I had taken a second look to see if there was something limiting the processors. I did not see anything.

I'm use Vista SP1 Home Premium 64 bit
Motherboard is the EVGA S58 SLI

I've attached images of the CPUz and Realtemp 3.0, had tried the beta version but showed only 2 cores also.

Regards

 

[FilipM]

Are my sensors bad? I know Core one reads higher than core 0, so on which one do i fall back? I have calibrated them once and I came up with 0 on Core 0 and -4.0 on Core 1. But Core 1 still reads hotter under load even though they are reading the same under idle.

 

[unclewebb]

Randallizer: If you look at CPU-Z it shows that you have only 2 cores and 4 threads. That's the reason why RealTemp is only showing temperature information for two of your cores. With a Core i7 it should show either 4 cores and 4 threads or if you have hyper threading enabled in the bios then it should show 4 cores and 8 threads.

You have a problem with your bios or how you have Windows set up. If your bios is set up correctly then try running msconfig and have a look in the Advanced Boot options.

Usually you don't have to mess with that but you could try forcing it to the largest number of cores available.
The 0123 for APIC ID in the RealTemp Settings window shows it is only seeing 4 threads. Did you do a fresh install of Vista when you built this computer?

File_1993: All 45nm Core 2 sensors are not great at reporting accurate temperatures. They weren't designed for this purpose and it shows.

What I've found in my testing is that TJMax is not a fixed number. It's not the same from one E8400 to another and the shocking part is that it isn't even consistent from one core to the next on the same CPU. Most of the 65nm Dual Core CPUs were pretty close from core to core but there are some 45nm sensors where the difference is 10C from core to core.

Try setting your idle calibration both back to 0.0 and 0.0.

Use TJMax = 100 / 107 for your two cores.

Now from idle to full load your cores should track each other very closely.

100 / 107 is my best guess on the data you've shown me. I took my Q6600 up into the 70C to 90+C range and there was a constant difference of 5C. I replaced the heat sink and the paste and etc. and finally concluded that TJMax is not consistent. For my Quad I use TJMax = 100 / 100 / 105 / 105.

Once you've set TJMax like this, try and run your CPU at low MHz and low core voltage and compare your results to your room temperature as outlined in the docs. The slopes of your two temperature curves are very similar. It's mostly just a difference in how TJMax is calibrated. Using the calibrated word on these sensors is kind of stretching it.

 

[Randallizer]

It is kind of embarrassing. I had checked what you had indicated and sure enough under advanced boot options it had shown 2 cores. Now I did not touch anthing there but my brother told me he had tweaked the system so that Vista will load faster. He had read that Vista used only 1 core to load and by changing this Vista will load faster so he had put to use 2 cores.
After putting it back to default all is now fine. I see now the 4 cores.

Thanks again as I would have probably RMA the cpu without your intervention.

Regards

 

[FilipM]

So that is 107degrees on Core 1 right. OK, I will report back when im done calibration for it

 

[unclewebb]

Randallizer: I'm glad I could help. That's an old tale on the internet about "how to speed up your boot times." I'm glad you've got a whole CPU now instead of just half of a Core i7. Now go give your brother a punch.

File_1993: What I like to do is run Prime95 Small FFTs for a couple of minutes, stop it, let it cool down for 30 seconds and then start it again a few times like this and watch how the two cores track each other. Having the logging feature on at a 1 second interval helps when you go back and look at the data.

When I do this with my E8400 the two cores are almost mirror images of each other and rarely vary by 1C. I think adjusting TJMax is going to get you a lot closer to accurate temperatures.

Here's is some of the calibration rge did over on XtremeSystems:
http://www.xtremesystems.org/forums/showpost.php?p=3384504&postcount=2429

Gathering data by drilling a hole in the top of your CPU to mount a thermocouple is not really recommended but I'm sure glad he did it.

Intel has admitted that there is some differences in TJMax but have not been willing to come out and say exactly how big this error can be. Most users would be shocked and the RMA line, or unhappy customer line would become way too long.

For the CPUs I've tested, it's usually about 65C or 70C when the two cores will start to move equally with each other. If there is a 5C or 7C difference here, there will almost always be the exact same difference all the way up to TJMax when one core will cause the CPU to start throttling. At this point, the difference in temperatures might change but from 70C to 97C, two cores side by side running Prime Small FFTs will show a consistent difference. The actual temperature is exactly the same for both cores. Any reported difference is sensor error.

If you're good with Excel you can graph your data and you should see two very similar temperature curves but they will be offset by about 7C.

 

[FilipM]

Ok, Ive done the Prime 95 testing 5 times, and the difference at default (no calibration, TJmax default) was from 6 to 8 degrees, mainly 7. So with the post above and the calibration that RGE suggested, here's what i came up with:

This is at 16.7C ambient temprature

 

[HTC]

It seems i have my E8400 core 0 sensor ill calibrated

This is my previous CPU: a lapped E6850, using a Noctua NH-C12P heatsink.

And this is my E8400: non-lapped, and using the same heatsink as above. The RAM has changed and increased too.

As you can see, core 1 sensor is stuck but what i'm intrigued about is the higher temp of core 0, when compared to the E6850: this has been so since day one and with the same version of Real Temp too (i just don't have a pic of it with version 2.84 of Real Temp).

@ first i thought it was because of one CPU being lapped and the other not but, though the ambient temp has increased, the minimum temp of the E8400, @ idle, has not, even though it doesn't seem to be stuck: the lowest i have seen it was @ 30º but it fluctuates between 30 and 32, on idle (C1E enabled).

Here's a pic of Real Temp taken a couple of minutes ago, with my settings:

Maybe you can help sort this out.

 

[unclewebb]

File_1993: I apologize. I screwed up. I reversed the 7 degree difference. What I meant to say was

Core0 TJMax = 107
Core1 TJMax = 100

Can you try doing a Cool Down Test at 1600 MHz? I just want to make sure that both your sensors are still moving at this point. Rge's testing was with his case open so try to do the same when comparing to his numbers. A closed case can add a few degrees to your results so it's best to open it up to be consistent

With the 107/100 TJMax numbers, these two cores are going to track each other very closely from idle to TJMax. Then you might need similar negative idle calibration factors if your sensors are not sticking during your low MHz test. Let your CPU sit idle at 1600 MHz for a couple of minutes before recording the temperature.

HTC: None of the sensors on any Core processor are 100% accurate from idle to TJMax. None of them. Not the original 65nm CPUs like your E6850 and definitely not any of the 45nm Core 2 based chips. With a 15C/17C Minimum temperature being reported, it looks like the E6850 sensors read too low and likely your E8400 reads too high. These sensors are not designed for accuracy at very low temperatures and a 10C error isn't unusual. If one CPU has 10 degrees of error one way and the other CPU has error going the other way then the combined error when comparing two different CPUs can be significant.

The best way to figure out what's going on is to test both CPUs under identical conditions, same room temperature, core voltage and MHz and then compare your reported temperatures to what rge found during his testing:

http://www.xtremesystems.org/forums/showpost.php?p=3384504&postcount=2429

I wish the sensors were more accurate than this but they're not.

 

[HTC]

HTC: None of the sensors on any Core processor are 100% accurate from idle to TJMax. None of them. Not the original 65nm CPUs like your E6850 and definitely not any of the 45nm Core 2 based chips. With a 15C/17C Minimum temperature being reported, it looks like the E6850 sensors read too low and likely your E8400 reads too high. These sensors are not designed for accuracy at very low temperatures and a 10C error isn't unusual. If one CPU has 10 degrees of error one way and the other CPU has error going the other way then the combined error when comparing two different CPUs can be significant.

The best way to figure out what's going on is to test both CPUs under identical conditions, same room temperature, core voltage and MHz and then compare your reported temperatures to what rge found during his testing:

http://www.xtremesystems.org/forums/showpost.php?p=3384504&postcount=2429

I wish the sensors were more accurate than this but they're not.

I thought it was possible for those temps on the E6850 since my house doesn't have AC and it was quite cold then, with ambient temp @ about 10º (i have this case).

What surprised me was that, when i placed the E8400, the temps were 15-20º higher, using the same CPU voltage and, @ the time, the same RAM.

The lowest temp ever on the E6850 (that i saw) was 15º on both cores with core 0 reaching that temp and staying there but with core 1, usually, with 4-6º more and spikes down to 15º.

Both CPUs have identical readings while running Linpack (i mean 1 core VS the other: not 1 CPU VS the other).

I'll try to use the same settings as the E6850 CPU for this but i won't be able to do it until this weekend. Hopefully, this might help to calibrate both CPU sensor readings.

 

[FilipM]

Ok Webb, I will try that tomorow, including the cooldown at 1600Mhz. I did the calibration in my previous post at 1600Mhz, 1.1V and an opened case.

 

[shaft103]

Hello. I was hoping to get some assistance in calibrating real temp before I start overclocking this system. I have attached an image of the Cool Down Test for your review. Your help is greatly appreciated!

 

[Kursah]

Shaft are you on stock cooling? I would say those temps reveal a stock cooler imo, with a cheap AC Freezer 7 Pro you'd be quite a bit less then that under load or should be. The cores could be reading higher though. I OC'd an e6400 in a friends rig to 3.2GHz @ 400FSB (8x400), it took 1.28v (CPU v under load as read from Everest, CPU-z, DFI's Smartguardian) to be orthos stable and peaked around 48C. It'll hit around 54C when it gets warmer out, but case airflow and cooling are vital before overclocking. That chip has some room to give on temps, but just wanted to say those look a bit warm for stock to be doing much for overclocking...just keep an eye on them, i'd say keep them below 65C per core.

 

[FilipM]

Did all of the wanted, here's how it looks now:

cooldwon test at 1600Mhz, 1.1V, open case, 19C

Calibrated, 1,1V, open case, 19C

 

[shaft103]

Shaft are you on stock cooling? I would say those temps reveal a stock cooler imo, with a cheap AC Freezer 7 Pro you'd be quite a bit less then that under load or should be. The cores could be reading higher though. I OC'd an e6400 in a friends rig to 3.2GHz @ 400FSB (8x400), it took 1.28v (CPU v under load as read from Everest, CPU-z, DFI's Smartguardian) to be orthos stable and peaked around 48C. It'll hit around 54C when it gets warmer out, but case airflow and cooling are vital before overclocking. That chip has some room to give on temps, but just wanted to say those look a bit warm for stock to be doing much for overclocking...just keep an eye on them, i'd say keep them below 65C per core.

I'm actually not on stock cooling. I'm using the Xigmatek HDT-S1283. I wonder if maybe I need to re-seat it because I also was expecting better temperature performance. When I was on stock cooling with my old mobo and this chip, I had it up to 3.2ghz with temps right around 60C.

 

[FilipM]

May I ask, cos I am seeing in loads of forums, to keep these processors (e8 series) under 60 deg at max load, where as on the intel site, it says thermal specification 72.4C deg. So 72.4 is the max safe and 10x is the thermal junction?

 

[FilipM]

Webb, instead of putting Core 1 with 107TJmax, would it be better if i put Core 0 on 100deg, and Core 1 on 93?

Something like this:

15C room temp, opened case, 1600mhz, 1.1V