10 Feb
2010
10 Feb
'10
9:50 p.m.
On Wed, Feb 10, 2010 at 3:15 PM, Roger Holmes
<roger.holmes at microspot.co.uk> wrote:
I would like to discuss Moore's law and how it
seems to have broken down in recent years. Processor speeds are still increasing but not
at the expected rate, but I wonder if the real problem is RAM speed, which does not seemed
to have kept up, and no longer seems to be quoted when you buy a computer
Even allowing for "main memory" moving to on-chip cache, of all of
those "500X over 25 years" multipliers I was mentioning yesterday, RAM
speed was not one of them. 250ns DRAMs were available commercially in
quantity 25 years ago, as were, IIRC, 200ns. Faster, smaller, more
exotic RAMs were also available, but didn't tend to get used for main
memory. Someone who knows Crays better than I can confirm, but ISTR
the Cray-1 used 150ns chips with something like a 32-way interleave.
Screamingly fast until you hit a branch.
Today, 7ns memory is common as muck, but 500 picosecond memory is not
(which would be about right for 500X the speed of memory from 25 years
ago). Of all the things that have gotten faster by leaps and bounds,
I/O and memory buses have made a fraction of the gains of raw clock
speed and storage densities. We are still much better off (compare
Unibus to e-cheapo desktop PCI), but not quite on the same scales.
Here's a prediction from 2000 about machines today...
http://www.ausairpower.net/OSR-0700.html
-ethan