A system running more hardware will, for all practical purposes, use more energy.  What this would do is increase the efficiency of that power use.  Say you're using a single threaded indexing program, and its indexing a very slow medium.  Why use a CPU processor when you can idle them, and idle most of the other GPU processors and just use the one?  This is mainly for max hardware utilization though.

In the VERY long run, I'm seeing thing trending towards very distributed models.  As system resources grow, I believe it will become practical to "network" within a system.  This can manifest itself in two ways.  First, is that due to multi-core systems slowly changing to many-core systems, a networking model is very scalable and with so many things to break, the fault tolerance will become a must.  This could allow then for computer systems to continue their march towards a more biological like organization, like a multi-cellular organism.  This will likely be abstracted to programmers and users, but on a hardware level, it allows for variable redundancy, extreme fault tolerance, internal and external networking models, and any few components which break will have no or minimal impact on the stability and usability of the system.  This is WAY WAY in the future, but that's where I imagine it going and this could be a step in that direction.  Was that as coherent as it should be?  I'm still playing with this in the back of my head, so its by no means well planned :P  I'd be more than happy to talk to someone about this, because no one at my university knows this area--our math and CS/CIS departments are feeble.

On Mon, Jun 6, 2011 at 9:19 PM, erik quanstrom <quanstro@quanstro.net> wrote:
> Well, two reasons come to my mind immediately.  First, I'd be cool.  Second,
> the wattage you listed is the max wattage, not the idle or light load
> wattage which would likely be used.  Per processing element, GPUs use less
> power, and you get more processing power per watt than a CPU under certain
> loads.

i'd sure like a reference to a case where a system with a gpu draws less
power than the same system without.  it's not like you can turn the cpu
off.

> This concept could be taken as far as to bring all processing off
> specialized areas for general purpose use, allowing potentially for an
> internally distributed system with high regularity, fault tolerance, etc.
> That's on the far end, but not to be totally discounted.

please explain.  how is a machine more of any of these things than
a regular multi-core machine?

- erik