Page 1 of 1

The Tjunction myth. lets come to a final conclusion

Posted: Mon Apr 02, 2007 7:33 am
by GGuyZ
The coolest, I have much appreciation for you and your developments, but I feel that certain issues must be solved, or atleast clarified as soon as possible. I'd appreciate it if you could share your conclusions with us in here, and comment on what I'm about to say.

According to one of the more interesting threads I've read in Intel's open Dev forums, a very unpleasant picture arrises. Desktop/Server CPUs DON'T have a constant Tjunction temperature, unlike many Mobile CPUs which contains the Tjunction information in bit 30 of MSR 0xEE. This value is either 85/100. Possibly by a mistake, this value seems to exist in the desktop processors as well, but this does not make it valid at all.

To make things short, there is no real way to predict an absolute temperature reading in C2Ds CPUs.

The Coolest, I'm sure you're aware of all the above, which makes me wonder why did v0.95 change the L2 revision chips to have a 100c Tjunction, while you know that this value is meaningless?

If you believe these chips tend to get hotter(well, much hotter), then would you please mention your knowledge of why this is true? I can't see why a newer stepping will get 15c hotter on the same archictecture.

The different conclusions don't seem to add up. Intel does not supply much support on this subject, and it only seems to contribute to the mass confusion.

Your response will be highly appreciated.

Keep on the good work,
Guy.

Re: The Tjunction myth. lets come to a final conclusion

Posted: Wed Apr 04, 2007 3:00 am
by Rhialto
GGuyZ wrote:To make things short, there is no real way to predict an absolute temperature reading in C2Ds CPUs.
What about Intel TAT application? I see no warning readings could be false...

Posted: Wed Apr 04, 2007 1:10 pm
by GGuyZ
TAT was developed for mobile CPU usage. The whole Tjunction minus DTSDelta was good for these CPUs(with small deviations of several degrees above or below), but it is basically useless for desktop CPUs. In desktop CPUs the calculation is PROCHOT# minus DTSDelta, and obviously we cannot read PROCHOT# temp.

The Coolest, I'd really appreciate your reply regarding this.