Y
J
johnno
Tom's power measurements are for full system power, not just CPU - so
implying that the 17% decrease in system power means thermal density has
increased is misleading at best. I think it's more interesting to look at
the increase in power between idle and full load. For Cedar Mill this is
55W, and for Prescott 86W. So that's about a 36% decrease in active power
for a 33% die shrink, meaning themal density has actually decreased a
little. Of course this analysis is almost as simplistic as the other, but I
suspect it's somewhat closer to the truth.
B
Brett Davis
johnno said:
Wrong
You did not read the article, they used the same motherboard and swapped
the 90nm P4 (Prescott) for a 65nm P4, (Cedar Mill) so ALL of the power
difference is caused by the CPU.
17% less power on a chip 33% smaller means the power density has gone
up, the new Cedar Mill will ship at slower speeds than existing P4s.
Looking at the difference between idle and load is silly, but even that
caps Cedar Mill at the same clock speed as Prescott. Intel has been
stuck at 3.8GHz for something like two years, longer if you count the
time stuck at 3.6GHz, normally clock speeds would have roughly doubled
over that time span.
Also Intel is going dual core, so the clocks will be WAY less than what
users are used to to stop the chips from melting. Downgrading to 2.8GHz
is not what I call an upgrade, thats a Celeron "value" chip speed, and
that is the "fast" one, how about 2.4GHz or less. (AMD's dual core chips
are available at the same speeds as the single core chips.)
I can't see any scientist or researcher with a brain buying an Intel
based computer when AMD based machines are available.
Paxville vs Opteron:
http://www.gamepc.com/labs/print_content.asp?id=paxville
The comparisons are grim for Intel.
Today comp.arch is all about managing thermal density, or Moore's Law
will end.
Brett
Flamebait: Maybe I should add a newsgroup ref for comp.arch.dead
K
Ketil Malde
Brett Davis said:
Not only today, but yesterday as well. If you look at the history of
CPU frequencies, you'll see that "doubling every two years" was a long
time ago, and only upheld somewhat by the P4 performing less work per
clock than alternatives.
-k