[GRLUG] Solid State Drives

[Tim Schmidt]

On Tue, Dec 2, 2008 at 11:00 AM, Michael Mol <mikemol at gmail.com> wrote:
> Oi...Disregarding CPU consumption, that would massively complicate the
> swap system.  Your memory pages are no longer a fixed size on disk,
> and may even grow beyond their initial size with nasty enough data.
> So you'd have to use more disk space to guarantee holding the same
> volume of original data.

Yup.  Such are the trade-offs one becomes willing to make when one's
CPU(s) are several billion times faster than their associated mass
storage.

Eventually, all this will change of course...  as the silicon black
hole that is the CPU continues to devour more and more discrete
components.  Co-processors have succumb, as have memory controllers
and fast local buses (Intel's Quickpath, AMD's Hypertransport, and
soon PCI Express), GPUs are next.  As bus widths widen (Intel's Core
i7 has a built-in tri-channel (192bit wide) DDR3 memory controller)
and speeds increase, the motivation to integrate the connected
components onto a single die increases.  Ram will likely circle the
void for some time, due to consumer's familiarity and acceptance of
the ability to 'upgrade' ram without changing other components of the
computer, but the writing is on the wall.  Ram will eventually make
it's way onto the CPU die, and become just another feature on which to
base a purchasing decision.  "The Intel XXX comes with 4Gb ram on die,
the XXX^2 costs $100 more, but comes with 8Gb."

As more components are gobbled up by the CPU's die, relative
performance of those components will change drastically, and we'll be
re-thinking relationships again.

--tim