CPU speed slowing down

According to Cadcamnet (Wolf publications), you won't be seeing huge speed
increases in CPU's over the next several years (at least with Intel). Part
of this is due to the fact that Intel doesn't consider CAD/engineering
software use as being statistically signifigant with regards to market
share. They see the current fastest CPU's as being more than adequate to run
typical office, consumer, and games software. In order to make faster chips,
Intel will have to develop and build new fabrication equipment based on .09
micron features. Given their 15-20% sales slump, they've decided to stick
with the .13 micron process for the forseeable future. AMD already has a .09
micron line for their Opteron 64 bit chips, so any performance inhancements
will probably come from them, but word has it that they will slow down speed
increases as well.

So, what does that mean to us. For one thing, if you currently use the
fastest Intel CPU, you'll just have to get used to SW pathetic performance,
and the 10-15% decrease with each new release. Or you can look for
alternatives.

It also means that SW won't be able to continue its lame practice of
claiming performance increases based on the latest fastest hardware. They'll
either have to clean up and speed up thier code, or lose customers.

Regards

Mark

 

I don't know about Cadcamnet, but Anandtech has a report today to the
contrary.
http://www.anandtech.com/cpu/showdoc.html?i=1948
This is Intel's "roadmap" for this year. In Q1, '04, they will have a 90nm,
P4, 3.4ghz, 1mb cache unit and by Q4, '04 they will have a 90nm, P4, 4.0
ghz, 1 mb cache piece on a new socket called LGA-775. Seems like a fair bit
of advancement to me.

 

Microns bring you more GHz and more transistors.
You need more transistors for 64 bits.
64 bits brings you essentially more RAM, not much more power for CAD
(floating point).

But you also need more transistors for dual cores processors, which will
bring you true dual processing at lower cost and without the FSB bottleneck.
*IF* SW was preparing its code to take advantage of multiprocessing
(parallelisation, thread-safe, ...) it *COULD* explain some
speed/reliability issues in the latest versions and *WOULD* open a brighter
future...
Just my 2 eurocents guess...

 

branches?

Check your parts FMT (which is actually a graph) with www.cadml.com ;-)
There could be some parallel branches, but it's not enough to process them
in parallel.
Think about a cube with 2 holes on different faces that intersect in the
middle of the cube :
the FMT show them as parallel, while the faces actually depend of each
other...

I believe parallelism can be exploited at an even finer grain, for example
to compute faces in shells.
But these are more related to Parasolid than SW.