Real Temp and Core Temp, not to be taken seriously

[spirits247]

From what I've been reading, people get themselves in a right twist over DTS/core sensor readings from CPU's. It got even worse when 45nm core2's were released as some of these sensors varied or were stuck at low temps. People were sending back CPU's over differences in DTS sensors and these programs are to blame for not making things clear.

From what I can tell, DTS sensors are only used to trigger throttling and shutdown for each individual core. They are not calibrated to give any kind of accurate reading across temp ranges, so their values are of no importance. They are factory calibrated individually per core to throttle at a point before damage or errors creep in. Not every core is calibrated exactly the same or gives out the same readings, but each core will throttle and shutdown correctly so no point analysing these sensors.

The only temp reading that matters and the one Intel base all their requirements and max limits on is the Tcase sensor - and that should always be below say 70C (for some core2's, depends on CPU type) under load. Usually this is the one the motherboard reads.

No matter what the DTS sensors are reporting, keep Tcase under the specified value set by intel and your CPU will work fine. The CPU will hit Tcase max way before any DTS sensor triggers throttling regardless of what reading they are churning out. So why all the obsession with core/DTS readings?

 

[Justintime]

The temperature reading is a guestimation via software going by what the average Tjmax of the cpu is, however the countdown to Tjmax done by the DTS is important, not the actual temp reading though that gives an idea how your cooling is working, Tcase is the temp as read at the IHS and core temp will be about 10-15C higher so for a Q6600 with Tcase MAX of 71C or whatever it is, 81-85C should be fine and i've dealt with tens of them to know those temps don't faze them one bit. I usually stick with 20-25C max away from Tjunction max for peace of mind, so for a i7 920 i'd say 80-85C max (linx, ibt etc.) loaded is acceptable etc..
Where did you get the Tcase being only 60-65C? It varies with cpu. You're basically saying the opposite of what most people actually reccommend.

 

[spirits247]

I agree, the countdown is important but even Intel say that there is no set TJmax for a particular core - only a guideline as to where it may be - it can't be read from the chip directly or calculated absolutely and does vary from core to core probably due to different voltage leakage, imperfections, variations...

...so even the countdown to TJmax isn't accurate and might be 5-10C off the real throttle number if a program assumes a set tjmax for a given CPU core even at fully loaded temps. Plus the sensors slope all over the place, even more so at lower temps, only getting accurate/triggering at a pre-determined throttle value set at the factory (and not readable from the chip) that can vary from core to core.

It's all guess work using these core sensors. All we need to know is they will do their job when it matters, and forget about them.

Realtemp has calibration routine instructions, which helps a little, but I still think trying to use these sensors for something they were not designed for is madness. Plus hardly anyone goes through the calibration routine.

If the CPU is working fine that's all that matters and the way you assess what's fine is the temp at the centre of the CPU's IHS (the Tcase or bios temp) and not the inaccurate and guesswork core/DTS temps. Even this isn't accurate without using a proper external thermometer, but it's a better guideline and what Intel states is important.

You are right however and I accept these programs do have a good use - they can be used as a very rough guide to seeing if your heatsink and paste is applied correctly since they are read using exactly the same software and hardware, plus everyone posts them so you can see if something is drastically wrong.

Just wish we could get back to discussing Tcase temps also as the more important ones as that's what decides if you are in spec. Even if Tcase/bios temp is off by quite a bit due to different motherboard and bios implimentation, by the time it approaches the max temp stated by Intel for your CPU, you will still be a little off triggering the DTS/core sensors and throttling unless something is horribly wrong with the Bios. So the DTS/core sensors really don't matter and the Bios/Tcase one does.

 

[Justintime]

I agree, the countdown is important but even Intel say that there is no set TJmax for a particular core - only a guideline as to where it may be - it can't be read from the chip directly or calculated absolutely and does vary from core to core probably due to different voltage leakage, imperfections, variations...

...so even the countdown to TJmax isn't accurate and might be 5-10C off the real throttle number if a program assumes a set tjmax for a given CPU core even at fully loaded temps. Plus the sensors slope all over the place, even more so at lower temps, only getting accurate/triggering at a pre-determined throttle value set at the factory (and not readable from the chip) that can vary from core to core.

It's all guess work using these core sensors. All we need to know is they will do their job when it matters, and forget about them.

Realtemp has calibration routine instructions, which helps a little, but I still think trying to use these sensors for something they were not designed for is madness. Plus hardly anyone goes through the calibration routine.

If the CPU is working fine that's all that matters and the way you assess what's fine is the temp at the centre of the CPU's IHS (the Tcase or bios temp) and not the inaccurate and guesswork core/DTS temps. Even this isn't accurate, but it's a better guideline and what Intel states is important.

You are right however and I accept these programs do have a good use - they can be used as a very rough guide to seeing if your heatsink and paste is applied correctly since they are read using exactly the same software and hardware, plus everyone posts them so you can see if something is drastically wrong.

Just wish we could get back to discussing Tcase temps also as the more important ones as that's what decides if you are in spec.

The only problem is, are Tcase temps dependant on the Bios and/or a separate hardware chip (thus making comparisions harder)? Or can we monitor Tcase directly from the CPU like people do now with the core/DTS sensors?

I think they depend on Bios implimentation, but it should be a reasonable guide as to if the CPU is working in an acceptable temp range vs Intels specs.

Tcase as read from the motherboard sensor is actually less accurate than read from the core due to the sensor being 3rd party and its accuracy highly dependent on its implementation/position and can and does vary even with BIOS revisions. What Intel means when they say there is no set Tjmax is that even though the program assumes a Tjmax, its not necessarily the Tjmax of that core, so for example, if the program assumes Tjmax is 100C and the sensors read 15C to Tjmax the program will assume 85C, however if the Tjmax is actually 105C for that particular core, then the actual temp is 90C, they don't mean the DTS is inaccurate, the 15C to Tjmax is more or less accurate and the throttle is dependent on that very DTS reading, theres no other sensor for Throttle or Tjmax, its just the program used to read it will display a wrong temperature, no biggie. Also theres no way to reaf IHS values straight from the cpu itself, it has to be read by 3rd party sensor.

 

[spirits247]

Where did you get the Tcase being only 60-65C? It varies with cpu. You're basically saying the opposite of what most people actually reccommend.

My mistake - I corrected it thanks for pointing that out. Yes, varies from CPU to CPU. My core2 was 71.2C or something, that's where I got keep below 60-65C bios/Tcase temp.

 

[Justintime]

BTW the good news is, i had some communication with the author of Realtemp (basically emailed him about a sensor issue i had with a 920 cpu) and he reckons the i7 cpus DTS implementation is much more superior/accurate to Core2 based cpus.

 

[spirits247]

Tcase as read from the motherboard sensor is actually less accurate than read from the core due to the sensor being 3rd party and its accuracy highly dependent on its implementation/position and can and does vary even with BIOS revisions. What Intel means when they say there is no set Tjmax is that even though the program assumes a Tjmax, its not necessarily the Tjmax of that core, so for example, if the program assumes Tjmax is 100C and the sensors read 15C to Tjmax the program will assume 85C, however if the Tjmax is actually 105C for that particular core, then the actual temp is 90C, they don't mean the DTS is inaccurate, the 15C to Tjmax is more or less accurate and the throttle is dependent on that very DTS reading, theres no other sensor for Throttle or Tjmax, its just the program used to read it will display a wrong temperature, no biggie. Also theres no way to reaf IHS values straight from the cpu itself, it has to be read by 3rd party sensor.

I read that there is a sensor - in the CPU itself - for approx IHS temps positioned between the cores - intended for the purpose of rough Tcase temp readings. It's this what determines if to speed up fans for instance via the bios and or monitoring chip.

Long gone are the days of external motherboard thermistors.

But yeah, you do require external implimentation by the motherboard to use it and yes it can vary from board to board, but that's the only one I take much notice of.

I was just pointing out that people stress over differences and readings that these core/DTS sensors sometimes give, but if they are doing the job they were designed and calibrated for, that's throttle and shutdown, then they are fine and just forget about them.

 

[spirits247]

BTW the good news is, i had some communication with the author of Realtemp (basically emailed him about a sensor issue i had with a 920 cpu) and he reckons the i7 cpus DTS implementation is much more superior/accurate to Core2 based cpus.

Nice to know - probably because Intel had so many peeps complaining about their CPU's based on core/DTS sensor readings using Realtemp, so they have had to respond by making them better!

 

[gareth170]

Where did you get the Tcase being only 60-65C?

amd chips?

 

[spirits247]

the 15C to Tjmax is more or less accurate and the throttle is dependent on that very DTS reading, theres no other sensor for Throttle or Tjmax, its just the program used to read it will display a wrong temperature, no biggie.

I must admit, I did get the way it reports data slightly wrong in my explanation and you've made me realise how it works - the distance to TJmax is an accurate countdown, directly from the DTS, even though we don't know what TJmax is for certain so we can subtract to calculate the actual real temp.

However, the scale of the countdown is not accurate across the whole temp range as it's not a linear sensor. It's only accurate very close to the throttling temp. So if one core reads 42C to TJmax while the core next to it reads 52C, these values are not accurate temps to TJmax, just a non-linear countdown and can't be compared. The actual countdown temps for both cores are unknown.

Moving on to really high temps, when the CPU is at 90C for instance, DTS (distance to TJmax) might read 10C, this number now being much more accurate and will throttle within 10C as it closes on zero/TJmax. The other core may now read 12C, again a more accurate reading.

The IHS sensor is an analog linear sensor, over the whole temp range, so as long as you are getting some sensible looking readings and it's a good bit lower than the thermal limit imposed by Intel (a IHS limit), the CPU should be working fine. You are hoping the motherboard maker has done a reasonable job.

Just let the core/DTS sensors do their thing, throttling and shutdown, as the stuck and differing values across cores are meaningless at low/medium temps, made worse by varying TJmax values.

In my experience, when the lowest DTS sensor is reading a lower value (like 20-30C to TJmax) the bios sensor is always showing a high value, like 60-70C. The bios temp usually always hits Intels limit before the DTS countdown limit is reached, so to me it's the better indicater of good operating temps as people (including me) make wrong assumptions when using Realtemp and Coretemp, thinking there is a problem with the CPU.

If people used it simply to see if they are too close to TJmax/throttling, where it's getting more accurate, that would be a good use of Realcore I guess. It's just the confusion it causes was my main issue and point.

 

[Justintime]

In my experience, when the lowest DTS sensor is reading a lower value (like 20-30C to TJmax) the bios sensor is always showing a high value, like 60-70C. The bios temp usually always hits Intels limit before the DTS countdown limit is reached, so to me it's the better indicater of good operating temps as people (including me) make wrong assumptions when using Realtemp and Coretemp, thinking there is a problem with the CPU.

If people used it simply to see if they are too close to TJmax/throttling, where it's getting more accurate, that would be a good use of Realcore I guess. It's just the confusion it causes was my main issue and point.

Odd, i find all the boards i have personally will read the cpu temp via speedfan or Motherboard Application much lower than Realtemp or Coretemp :

Gigabyte - DS3 965 - DS4 965 - DS3 p35, DQ6 X38
Asus - P5K-E Wifi, P5K-EPU, P5K-E, P6T Deluxe, P6T SE
+ A few socket P and Socket M mobile boards by Asus and Aopen

Can't say i've heard of the Tcase sensor on the cpu itself, that new to me as i always thought it was a motherboard sensor onboard that did that job.

 

[Justintime]

amd chips?

Honestly, haven't touched an AMD chip in that 'special way' since 939, those were the glory days Every AMD rig i've built since has been stock and office use, just briefly looked at temps and thats it

 

[spirits247]

Odd, i find all the boards i have personally will read the cpu temp via speedfan or Motherboard Application much lower than Realtemp or Coretemp :

I didn't put that in a very clear way - I meant the DTS/core sensor reads 20-30 distance to TJmax, which means Realtemp/coretemp calculates 70-80C temp reading after assuming a TJmax of 100C. The Tcase/bios temp reads around 60-70C at this time.

So what I meant was the Tcase/bios temp (even if not accurate) and Intels thermal spec are usually very conservative - you will hit 72.2C Tcase much sooner than a DTS sensor approaches 0 and throttles (meaning the core temp has hit TJmax).

Yes, core2's have a legacy sensor on the chip itself, as well as the DTS/core sensors.

Oh just ignore me - if it works fine in prime, you're good to go. Just decided to read up about all the sensor info around the net and these were my conclusions (which may be wrong). Just felt like going through all the info and letting others pick faults. Thanks for helping me out on stuff anyway.