New Intel specs, better temp monitoring

[mrmaigo]

Previously undisclosed information about DTS or digital thermal sensor by Intel will be announced at IDF to provide for more accurate internal temperature monitoring of current 45nm CPUs.

At IDF this year, Intel is expected to unveil (PDF) information that will lead to improved internal processor temperature http://www.tomshardware.com/news/IDF...ture,6154.htmlmonitoring. Intel has never previously disclosed information about the Tjunction value before of its 45nm desktop processors, a value that is needed to accurately measure the internal CPU temperature. Previously, these Tjunction values were estimated by software, but as each model and stepping of processor can potentially have a different Tjunction value, not to mention varying temperature response curves, no current method can be accurately trusted.

Old news? Still good news
-more-

[The Coolest]

Great news, and I like Anandtech's article better than Tom's. Tom's is inaccurate in several points.
http://www.anandtech.com/weblog/showpost.aspx?i=478

[Mumak]

Finally
But since DTS was never meant to be used for reporting of current temperature I'm wondering how will they manage the accuracy. The problem is that the sensor is most accurate only around the Tj,max (+/- 5 C accuracy) so it's suited for catching the hot point. At lower temperatures (normal usage) it's more inaccurate and the results may vary by +/- 15 C.

[thenrz]

Accurate temperature readings? Finally.

[K404]

I hope they fix the range/interpretation so they dont coldbug through misinterpretation

[zerotol]

so.. this should have been disclosed why dont i read anything about it then

[The Coolest]

Dunno...

[rosco]

The PDF is already available here : http://intel.wingateweb.com/US08/pub...TS001_100r.pdf

[The Coolest]

Thanks!

[Hornet331]

nice, now we know the TJmax for the QX9650 (95°C), which means Realtemp was right.

[The Coolest]

Never said it wasn't right, simply said that there is no real info from Intel to confirm this.
His tests were mostly conducted on C2Ds which are 100C.

[Praz]

Looking forward to the update of Core Temp.

[JumpingJack]

Great news, and I like Anandtech's article better than Tom's. Tom's is inaccurate in several points.
http://www.anandtech.com/weblog/showpost.aspx?i=478

Tom's is inaccurate in most of it's points for the past several years. Pathetic is a word that comes to mind.

[Praz]

Freecableguy has updated the MBM5 plugin with the proper 45nm CPU Tjunction values.

Motherboard Monitor 5 (MBM5) Core Temperature Plugin

[The Coolest]

That was pretty quick, heh.
It'll take me a day or two to get the time to do this for Core Temp.
I really wish they would release info on 65nm chips and Atom CPUs as well.

[AndrewL1950]

Hopefully Coretemp or RealTemp will be updated soon.

So did this let everyone know what the real Tjmax is and make a big difference between what we were using now?

[ken]

Hopefully Coretemp or RealTemp will be updated soon.

So did this let everyone know what the real Tjmax is and make a big difference between what we were using now?

well, not much apart from that my temp just went up by 5C :S

[AgentGOD]

I would love to know the real Tjunction on my Q6600!

[DragoonXX]

Interesting to see that the QX9770 has a TJmax of 85 degrees

[FischOderAal]

E8000/e7000: 100°c
Q900/q8000: 100°c
Qx9650: 95°c
Qx9770: 85°c

http://www.hardwareluxx.de/story.php?id=8133

[Hornet331]

unclewebb (developr of realtemp) nailed it pretty good:

You nailed it. This great announcement that was going to clear everything up has left a much bigger mess.

Using the correct TjMax will make your TjMax temps more accurate but it doesn't do anything to fix the rest of the temperature range. This has only shown me that the amount of possible sensor error in the low end is even greater than I originally imagined. I was giving these sensors more credit than they deserved.

MBM to release a quick update so that it is using the correct TjMax is pointless. If there is 20C of sensor error in the low end, and there is, then the correct TjMax doesn't mean squat except for the very rare times when you are at TjMax. Same thing for CoreTemp, Everest, SpeedFan etc. You can set the correct TjMax but if these programs don't give you a way to compensate for sensor error then the reported temps coming from them are not accurate.

Time to fire up Excel to see if I can come up with a better way to convert random data into some usable temperature numbers.



But even that simple statement isn't true. Intel's graph shows that the sensor error increases linearly with the distance from TjMax. In your example, the digital sensor reading would be 30 and that reading might already have 5 degrees of error in it. So the answer is 70C +/- 5C or somewhere between 65C and 75C. This announcement hasn't solved anything. My previous assumption was that the sensors are very accurate within 30C of TjMax but the Intel graph shows that sensor error starts at TjMax and continues linearly across the entire temperature range.

temp readings are worthless below tjmax, cause of the increasing error.

[Mumid]

unclewebb (developr of realtemp) nailed it pretty good:
temp readings are worthless below tjmax, cause of the increasing error.

After setting tjmax to 100 my core temps mirror my motherboard reported CPU temp to within 1c.

[makaka]

Yeah , this is my dream to see a 45 nm accurate temp , my current C0 cpu's all show crap and none of them show sub + 10 c temp

[Mumak]

And HWiNFO32 has been updated with new Tj,max values for Penryn and Nehalem families:
http://www.hwinfo.com/news32.html

[|Oc|REvO_HsArC]

this would make my e8400 5ºc hotter than realtemp shows? crap.... that would be well beyond 80ºc full load in my case. Has to be wrong. My water temp is only 30ºc

So to know the real temperature, I have to decrease my cpu mhz and vcore to the max and take a reading. The difference between my water temp and the c2d sensor readings is the value I have to subtract to the realtemp/coretemp (adjusted to tj100) ? correct?