[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Print to SHR

heuser.marcus@freenet.de wrote:

On Mar 7, 12:19 am, aiiad...@gmail.com wrote:


When did the 2.5Ghz come out?  It seems like there have been ZERO
speed improvements in the last ?5? years.

I have an article that I snipped out of the ?SanFrancisco Chronicle?
around 1999...  "Motorola introduces super-fast CPU" or something
like that. They claimed 75Ghz.  I'm pretty sure we had 2.0ghz
available at the time.



But now we have up to four cores in consumer CPUs and much
larger caches.

Intel just presented a prototype with 80 cores breaking the teraflop
barrier - but it's not x86-compatible and will stay in the drawers
(they use it to learn how to produce x86-multi-cores etc.).


And there's the rub.  Multi-thread performance is not the same
substance as single-thread performance.

Quoting peak parallel FLOPS is totally unrelated to the performance
that any user will see (who isn't solving partial differential
equations).  And "grid" computing parallelism, not shared memory,
is the wide open frontier of high-performance computing these days.


the hard disk guys continue
to out-run Moore with density and speed improvements.



I guess that's right but on the other hand Moore's Law seems
not to be some kind of "nature law" but some kind of "master plan"
for the silicon industry.


Right, which is why I always spell it Moore's "Law".

It is an economic plan which states that if you increase the
density and performance of chips fast enough, the market will
grow at a rate which will finance the R&D and capital investment
needed to keep improving.

That proposition is now seriously broken, in the sense that
the improvement has slowed, and so has the market growth.


Like "Why produce exponentially faster CPUs if this speedup
is all that is needed to sell them?"


And if marketing were all that was needed to float the industry,
then we could get rid of all the engineers.  ;-)

More seriously, the rate of growth of the chip industry has
taken a serious stumble as computers have become commodities.
Much of the "way of life" of Moore's "Law" has become unsustainable.


See: http://en.wikipedia.org/wiki/Moore's_law


Been there, done that, got a closet full of T shirts.  ;-)


It seems to me that reliability has gone down.  I haven't owned
a ton of the new big drives, but I have had 2 40GB, 1 80 GB
go bad on me in the last 4 years.



I on the other hand had very few harddisk defects and am
generally pleased with the progress in this field.
Hard disks get bigger, quieter, cooler, smaller & cheaper
all the time.


I agree--those folks are keeping the innovations coming,
with the Giant MagnetoResistive heads, the Partial Response
Maximum Likelihood read channel, and now vertical recording.


Graphics cards are nearly the opposite: They get faster
but bigger, louder, hotter and more expensive.


Higher clock rates * higher parallelism = much higher power
(= more noise and more $$).


If it weren't for massive consolidation and the cost reductions
that come from economies of scale, things would be even dryer in
silicon valley.


I read another article about speedup of CPU's...  The entire
motherboard circuitry was on a chip.  The chip was glued
to the board (non-upgradeable). There were fewer bus
speed restrictions, because the bus is in the chip.

I wonder when we'll get these?



I think I would welcome this if the price is right.

This sounds somewhat anachronistic for the modular PC world but
when I *really* upgrade my PC roughly every two years I'll have
to buy not only a new CPU & cooler but a new motherboard, new
RAM, a new graphics card and maybe a new power supply (maybe
even a new case).


The problem is coming up with a *single* congfiguration which
is both cheap enough for most and fully configured enough to
cover a large part of the market.  Fail to do *both* and you
lose the necessary volume.


Why is this? Because our friendly CPU supplies seem to
introduce new sockets faster than most people buy new underwear.
And this cycle accelerates, in my opinion.


This is primarily a result of the need for more memory bandwidth.

-michael

NadaNet networking for Apple II computers!
Home page:  http://members.aol.com/MJMahon/

"The wastebasket is our most important design
tool--and it's seriously underused."