[ltp] Re: Slightly OT: where are the 10GHz P6 thinkpads?

[Daniel Pittman]

Richard Neill <rn214@hermes.cam.ac.uk> writes:

> An interesting thought struck me today. For the last 8 years, I have
> upgraded every 2 years, in order to (roughly) double the performance of
> the computer. 2 years has now elapsed since my last upgrade, and it is
> time for another one. Yet there isn't any compelling new product to buy! 
> I wonder why that is, and if anyone can explain it?
>
> There's no obvious answer to this, either from intel,google or
> wikipedia. Also, the quantum limit should allow a 10nm process. So, why
> has Intel, in particular, stagnated?

Three big problems, heat, radiative noise and quantum mechanics, have
gotten in the way of delivering faster systems, and on delivering things
like a 10nm process, if you look at the current moves by all the big
chip manufacturers.

Radiative noise is a special killer: it hits home on longer stretches
outside the CPU, like the memory bus, where speed improvements are
extremely hard to achieve because of cross-channel interference and RF
emission problems.

Modern CPUs have outpaced memory performance significantly, to the point
that cache miss time usually dominates the performance for most tasks,
and improving that is a much harder job, in some ways, than improving
CPU performance.

> A datapoint: 2 years ago, the top desktop CPU was a 3.0GHz/800MHz
> P4. Now, the best available (at a sane price) is 3.6GHz. Similar things
> have occurred for laptops [admittedly, there have been gains in
> efficiency with the P4m, so the numbers are less meaningful] - and
> although my 4-year-old A22p (P3, 1GHz) could certainly use an upgrade, I
> wouldn't call it obsolete. It should be, though!

While you vaguely acknowledged it here, most of the current chips
deliver significant performance improvements at equivalent clock speeds
to the previous generations.  

They also deliver those accompanied by significant gains in
functionality, power use and heat generation, all of which require
significant engineering effort that could, otherwise, be spent of
improving raw processing power.


Finally, as you note, for many uses the machine they have is sufficient
for their needs.  Lower demand there, coupled with higher demand for
portable computing power, refocuses the industry to where the real money
is. :)

	Daniel