At the Intel Developer Forum this year, Intel disclosed some additional information regarding the Digital Thermal Sensors (DTS) on its 45nm processors, as well as the DTS in the upcoming Core i7 processors. Developers and enthusiasts have now had time to process the provided information, so how has this information been received?
Tom’s Hardware previously reported on the excitement surrounding the release of this information because it could finally put to rest the debate over which temperature monitoring software provides the most accurate readings. Unfortunately, Intel’s IDF presentation was quite disappointing, and once again the enthusiast community has been left scratching its head in the dark. While Intel did disclose the maximum Tjunction values for all of its 45nm desktop Core 2 Duos and Core 2 Quads as well as a few other small details, it stopped short of revealing any especially useful information.
According to the developer of RealTemp, “[Intel] did not release enough information for any software developer to write an accurate program so we’re right back to guessing and making assumptions. It’s easy to take a pen and circle some numbers but they didn’t test, prove or show anything.”
Of course, the primary function of the DTS is not for the end user to monitor temperatures, but to protect the processor from damage due to overheating. This makes it even more strange that Intel has released the Tjunction Max values for its 45nm processors.
These maximum temperatures are a nice bit of icing on the cake. Due to increasing inaccuracies in the readings taken from the DTS however, as the temperature moves further from this threshold (what Intel calls "slope error"), you can only assume decent measurement accuracy when you are running at a temperature equal to the maximum Tjunction value. This is extremely rare, because unless you did not mount your heatsink correctly, or you are running at extremely high voltages with insufficient cooling, you will probably never reach this temperature. RealTemp is currently the only software which attempts to offset these increasing sensor inaccuracies. Hopefully other software developers will follow suit.
But the problems don’t stop there. In fact, Intel stated in the presentation that Tjunction Max itself is subject to errors in calibration, and varies slightly from part to part. Also, Intel officially confirmed that the sensors can "bottom out" or "get stuck" at lower temperatures, something that several users have noticed for some time. The sensors are riddled with problems and this makes it extremely difficult to produce meaningful data from them.
“Intel’s presentation doesn’t give us any formulas or anything that anyone can agree on or use to create a formula to convert this DTS data into temperatures,” said the RealTemp developer.
For your reference, here are the official values that Intel provided in their presentation:
|E8000 and E7000 series||100C|
|Q9000 and Q8000 series||100C|
Unfortunately Intel did not disclose any maximum Tjunction values for their 65nm parts, which is both disappointing and strange. The 65nm CPUs precede its 45nm parts, and to us it seems logical to release information about prior technology rather than current technology. Perhaps Intel did not feel it was necessary since the 65nm CPUs have not had the same “bad press” as the 45nm CPUs with regards to erroneous DTS readings. While it is possible to produce reasonably accurate estimates of the maximum Tjunction temperature for the 65nm processors through testing and experimentation, it would save significant amounts of time if these details were already provided.
Looking at the rest of the information of interest provided in the presentation, we now know that the DTS in 45nm Core 2 processors are more prone to inaccuracies than originally believed, as slope error starts at the maximum Tjunction temperature where it was previously thought to be quite accurate within a certain range of this value. Intel also informed IDF attendees in the presentation that the DTS in the upcoming Core i7 processors will have range and slope improvements over current sensors.
So what does all this mean for enthusiasts? Essentially, taking the time to calibrate RealTemp correctly will provide a user with a "ball park" value that is reasonably close to their actual core temperatures, but without more information provided by Intel, truly accurate core temperatures across the entire temperature range simply isn’t possible. We can only hope that after the release of Core i7 Intel will divulge further details about its current line of processors. In the interim, users may wish to use the temperature that is supported by Intel’s Processor Spec Finder, known as "CPU temperature" or "Tcase", which runs about 5c lower than the Core temperatures, and is used by motherboard utilities, Everest, Hardware Monitor and SpeedFan.
Or we can forget all this and go back to the days where motherboards come with sensor pin-outs and we’d stick the thermal sensor wire directly onto the CPU.
Eh? Where is it? :(
As the author of the Core 2 Quad and Duo Temperature Guide - http://www.tomshardware.com/forum/221745-29-core-quad-temperature-guide - here at Tom's, I can fully appreciate the importance of having complete and accurate information from Intel upon which to base calibration procedures. I've been corresponding with Kevin Glynn since April, and I can tell you that we've both read hundreds of pages of Intel documents.
Although we've both discovered a few revealing papers, we were hoping that the substance needed to fill in the remaining blanks would come from Intel at IDF 2008 during the DTS presentation. Regardless, research and testing will continue toward finding the highest degree of accuracy, and with the help of articles such as this, hopefully Intel will eventually disclose Tjunction Max values for all Core 2 desktop variants.
Thank you Tom's Hardware for writing this follow up article on such a hotly debated topic. I'm certain that many users would find more temperature monitoring articles extremely helpful.
you know where to stick that thermometer! hehe