Re: temperature standard - global config option?

[H. Peter Anvin]

Followup to:  <[e-mail]>
By author:    Chris Boot <[e-mail]>
In newsgroup: linux.dev.kernel
> 
> Well then, tell all the teachers in this world that they're stupid, and tell
> everyone who learnt from them as well.  I'm in high school (gd. 11, junior)
> and my physics teacher is always screaming at us for putting too many
> decimal places or having them inconsistent.  There are certain situations
> where adding a ±1 is too cumbersome and / or clumsy, so you can specify the
> accuracy using just decimal places.
> 

This, again, is a presentation issue, and is irrelevant to the
intricacies of fixed-point arithmetric.

> For example, 5.00 would mean pretty much spot on 5 (anywhere from 4.995 to
> 5.00499), wheras 5 could mean anywhere from 4.5 to 5.499.
> 
> Please, let's quit this dumb argument.  We all know that thermistors and
> other types of cheap temperature gauges are very inaccurate, and I don't
> think expensive thermocouples will make it into computer sensors very soon.
> Plus, who the hell could care whether their chip is at 45.4 or 45.5 degrees?
> Does it really matter?  A difference of 0.1 will not decide whether your
> chip will fry.

Does it really matter NOW?  No.  However, 1 cK is a convenient unit
and a good use of bits.  Can we guarantee it won't matter in the
future, especially not on a CPU which may very well require complex
algorithms to eke out optimal performance in a thermally-challenged
environment (more than just simple trip points.)  Now it gets
interesting!  I have actually seen, in the lab, an algorithm which
required complex guesswork, because it required information below the
noise level of the sensor (and yes, it *is* possible to obtain that
information.)

	-hpa
-- 
<[e-mail]> at work, <[e-mail]> in private!
"Unix gives you enough rope to shoot yourself in the foot."
http://www.zytor.com/~hpa/puzzle.txt