Some questions regarding Tjunction values ...

[NoobyDoo]

Was just going thru "Intel's 45nm Dual-Core E8500: The Best Just Got Better" at AT. Need some answers to these noobish questions :

1) The article states "the MSR is designed to essentially count down the margin to the core's maximum thermal limit, often incorrectly referred to as 'Tjunction' " and "Intel representatives have informed us that these as-of-yet unpublished Tjunction values may actually vary from model to model - sometimes even between different steppings".

When is this Tjunction value actually decided/set ? Before/during/after the chip is fabricated ?

2) How exactly is the value in the single model specific register (MSR) set ? Is there some sort of temp sensing device within the CPU which reports the current temp, which is then subtracted from Tjunction ?
Speaking as a programmer, if :
msr_value = ( tjunction_value - current_temp_value )
both tjunction_value and current_temp_value must be known.

Guess what I want to know is how exactly is the mechanism of setting the MSR value actually implemented ?

3) Shouldn't users be more bothered about the value in the MSR ( Tjunction - temp ) rather than the absolute temps ?

Thanks.

 

[Mondoman]

Your answers are here: http://www.overclockers.com/articles1378/

 

[NoobyDoo]

Thanks. Answers Q2.

Q1 & Q3, anyone ?

4) Why doesn't Intel release Tjunction values ? Even if they "may actually vary from model to model - sometimes even between different steppings", the number of models & steppings is quite small. Wouldn't it benefit the end-users ?

 

[Mondoman]

1) AFAIK "Tjunction" is the temp reported by one of the in-core sensors, NOT the maximum such temp before throttling is initiated. The latter would presumably be called something like TjMax , and is determined by the chip design (before the chip is fabricated).

3) They need to worry about both, as while throttling behavior is determined by TjMax, electromigration increases with both temp and voltage, with no theoretically safe value for either.

4) Intel doesn't do everything that would benefit end users. The end users in this case are principally OCers; Intel would not seem to benefit by assisting their efforts.

 

[CTho9305]

the number of models & steppings is quite small.

That's unlikely to be true. Manufacturers only tell you about revisions when they have to (e.g. they made a change that may require different microcode patches or configuration values set by the BIOS) or see some benefit in telling you (e.g. power reduction, new feature). There can be plenty of revisions that are sold as the same part.