> I expect in a time to come there will be full clustering solutions on
> a single chip. It will be interesting to see how these are marketted
> to the home consumer, and how much bloat the software manufacturers
> are willing to add to make the hardware interesting.
>
Intel's behind the curve on this -- IBM (Cell processor), Sun (8 cores
x 4 threads/core per die in 1st-gen Niagara, 16 cores in 2nd-gen), AMD
(the crossbar architecture used by Opteron motherboards, in which each
CPU controls separate memory banks, and thus the memory layout is no
longer uniform) mean that "commodity" hardware is already encroaching
on what is traditionally the domain of computer clusters.
> My guess: Microsoft will create a new proprietary platform that only
> runs on the new hardware. All of their software will then be rewritten
> for this new platform. They'll stop supporting the old software and
I am not sure how successful Microsoft will be in this new market. The
last success they have with a new hardware platform is with
WinCE/Windows Mobile, and even that takes many iterations. X-Box 360,
which also has an interesting CPU architecture, is a loss-leader (MS'
gaming division lost, what, a billion dollars a year?). And Vista does
not even run well on dual-core laptops!
> Technology may change focus once we get to a point where rolling
> blackouts occur widespread throughout the US. I suspect this is
> just a matter of time. It wasn't that long ago in which we thought
> the idea of rolling blackouts occuring anywhere in the U.S. was
> an absurd idea. Now there's a (relatively) big market for home
> generators tied directly in to the wiring and kicking in
> automatically.
IBM's putting the Cell processor in workstations, I think. And blade
clusters are becoming more common. It's quite interesting how even the
server marketplace is becoming more interested in efficiency (due in
part to growing cooling requirements). I recall this recent study
that, interestingly, shows Opteron systems being more efficient than
Xeon equivalents if power usage is measured at the wall socket, but
less efficient at the CPU level, or if the system load is high.
Culprit? FB-DIMM memory controllers behave very poorly when idling --
their power requirement is mostly load-invariant.
--
Michel
_______________________________________________
BLUG mailing list
BLUG@linuxfan.com
http://mailman.cs.indiana.edu/mailman/listinfo/blug
No comments:
Post a Comment