The downside to Moore's law is that we are getting
substandard processor architecture.
Moore's law says that every 18 months computer performance will double. This means that every 3 months a computer's performance must increase by about 20%. Assuming that half of the increase in speed comes from improvements in
chip level process technology (smaller features, faster wires, better
VLSI design) then if you want to introduce a new improved method to, say,
predict branches and it'll improve processor performance by 5% then you have 45 days to not only get it done, but integrate it with the other billions of
transistors already in the processor and get the technology out the door.
Otherwise it's simply not worth it. If you can't improve processor perfromance by 10% in three months,
you might as well go home and cry, because the
AMD's of the world are
hot on your tail and they'll
eat your pie.
It's often difficult not to design the
technically correct solution. There
is a difference between the '
correct' way to do it and the '
quick and dirty' way. But the reality is that the correct solution is often not the right solution.
And that is why we have "fill in your favorite technically flawed hardware/software".