"Your thoughts on the above?" [Tcase=68 C for 2GHz, and 69C for 2.2GHz]
I totally agree with you, except throttling. AFAIK,
the unconditional throttling kicks in if the chip
temperature exceeds 90 (or 120?) degrees C, and the
set point is not programmable. So there must be some
disconnect in operations, between 69C and 90C: the chip
become non-operational before the thermal protection
is engaged. [there is a S/W-level of thermal throttling,
but it is slightly different story].
Also, Intel is setting a lower limit for operating
temperature, +5C, which is also unusual for CPUs.
I think Intel has problems with some circuitry design;
it could be an inherent problem with self-timed circuits,
but it is my WAG.
Athlons do not have this limitation, and can be chilled
down [theoretically<g>] to get higher performance.
P4 can't.
The 5% increase in power for 10% increase in frequency
is due to the simple fact of 17A quiescent current,
so the dynamical power increases just for the proper
fraction. However, I do agree with you that there is a
funny differentiation - Intel has announced the "lower 44W
thermal envelope" chips, at 1600, 1800, 2000. They have
lower leakage currents, about 10-12A. I think they are
just different bins from wafer sorts: low-leakage chips
are usually slower, and occasional high-leakage (but
still somewhat acceptable) constitute the 2.2GHz bin.
If this is true, it would be another indication of
limitations of current P4/0.13um technology.
However, this does not mean that Intel cannot fudge the
process and get somewhat better results in the future.
"Thanks for providing the link for the pdf."
Any time. However, I thought that for anyone who
watches the PC industry, this is a mandatory link
to visit frequently:
developer.intel.com
Regards,
- Ali |
