[extropy-chat] A slow down in Moore's law?

Alfio Puglisi puglisi at arcetri.astro.it
Mon Oct 18 19:07:37 UTC 2004


On Mon, 18 Oct 2004, Kevin Freels wrote:

>So would you say that processing speeds are still doubling every 18 months
>even though clock speeds are not?

Actually it's not that easy to say. Processing speed is now depending a
great deal on what type of computation you do. In general usage (internet,
mail, starting office and so on) the main bottleneck is often the hard
disk, not the processor or the memory, and sadly the hard disk performance
is not going up as fast as the rest (even if the capacity is). So you
won't see much difference.

For multimedia processing, current processors are much faster than, say,
4x with respect to 3 years ago, thanks to specialized instructions sets
like SSE. 3D graphics cards also have been on accelerated Moore's law for
some years now, but the type of processing that they do is very
specialized. Things like 3d rendering with 3DStudio or Lightwave have
benefitted from this too and are now faster than one should expect looking
at the clockspeed.

For scientific workloads, those measured by SPEC benchmarks and similar, I
believe things have progressed quite close to a doubling every 18 months.
But unless you run numeric-intensive programs like simulations you won't
see it.

Commercial benchmarks like TPC, useful for banks and such, depend a lot
more on system architecture than just the processors used.

So as you see it's quite complicated so say how processing speed is going.
The general direction is up, but the slope is hard to measure :-))

Alfio



More information about the extropy-chat mailing list