Re: Moore's law resuccitated

On Feb 1, 8:56 am, Phil Carmody <thefatphil_demun...@xxxxxxxxxxx>
wrote:
"r...@xxxxxxxxxx" <r...@xxxxxxxxxx> writes:
On Jan 31, 9:39 pm, Jerry Coffin <jcof...@xxxxxxxxx> wrote:
In article <1170227003.838465.108...@xxxxxxxxxxxxxxxxxxxxxxxxxxx>,
r...@xxxxxxxxxx says...

[ ... ]

The other important thing to note is that doubling doesn't occur by
simply shrinking the transistors. This is also achieved by developing
processes to handle larger wafers (and die). Chips today are *much*
larger than they were back then, which is one of the reasons (along
with shrinking transistor sizes) that we've packed so many transistors
on a single chip.

There's been virtually no growth in die size for _years_ now. Just for
example, in 1997 Intel was selling Pentium/MMX and Pentium II
processors. The Pentium/MMX was 129 square millimeters and the Pentium
II was 203 square millimeters.

"Years" is a relative term here. Today's dice are much larger than
they were in 1965 when Gordon noted his findings. And that's one of
the contributing factors to the exponential growth of the size of
semiconductors. It wasn't all just shrinkage.

You're right, die size definitely increased steadily over time
from the earliest microprocessors all the way to the mid 90s -
about the pentium pro era. The last decade has seen a slight
reversed that trend. However, increased multi-coring could buck
that reversal.

Interestingly enough, it was during that period (up to about 2000,
actually) that speed increases more or less matched Moore's law.
Performance boosts took a nap during the early 2000s (though the
recent spat of CPU releases have resumed the doubling of performance
-- e.g., my core2 duo is twice as fast as my PIV machine).

Of course, exponential growth (size or speed) cannot be maintained
forever. Though we've been hearing "Moore's law only has about 10
years left in it" for the past 15 years or so, we truly are
approaching the point where they cannot keep this up. And as they
scale the geometries down to the atomic level, the only way to push
Moore's law any farther is to use larger chips. That, of course, will
fail eventually too. Whether we have 10, 20, or 30 years left to
Moore's law, it's pretty clear we're in the last stages now unless
there is a phenomenal breakthrough in manufacturing processes. Even
then, that's only going to extend it a short period. Obviously,
expoential growth is unmaintainable. Everyone but the press realizes
this.
Cheers,
Randy Hyde

.

Relevant Pages

  • Re: Moores law resuccitated
    ... simply shrinking the transistors. ... There's been virtually no growth in die size for _years_ now. ... The Pentium/MMX was 129 square millimeters and the Pentium ...
    (alt.lang.asm)
  • Re: Moores law resuccitated
    ... simply shrinking the transistors. ... Chips today are *much* ... The Pentium/MMX was 129 square millimeters and the Pentium ...
    (alt.lang.asm)
  • Re: Tilera: how exciting?
    ... using a Wallace Tree multiplier. ... condition on whether or not a multicore chip is really interesting. ... faster than the increase in the number of transistors used, ... The original Pentium had 3.1 million transistors, ...
    (comp.arch)
  • Re: CAD and Pentium 4 vs Pentium M
    ... > list of specs says I need a pentium 4 chip with at least 3 GHz. ... > of them are still making notebooks with Pentium 4 chips. ... > performance in CAD as a 3.0 GHz Pentium 4 chip? ... Be sure to get at least 1GB RAM, and a separate gfx (no ...
    (comp.sys.intel)
  • Re: MAT 04
    ... offer some sort of bone-simple custom chips. ... I've had Japanese RF transistors that came as arrays in SIP packages and they were very behaved. ... those chips should behave. ...
    (sci.electronics.design)