I'm no physicist, but I did a back-of-the-envelope calculation (available if you are curious, my "envelope" is vi, but since I'm a lousy physicist I gloss), using http://en.wikipedia.org/wiki/Limits_to_computation (which connects the cosmic microwave background radiation to a limit for energy per state change) and guesstimating a million state changes per FLOP (I don't know how to count the state changes to perform e.g. a multiply of two registers, because a circuit to do it would have current or taut strings or something that themselves change state), and got 10^-15 joule per FLOP or 10^-15 watt per FLOPS, about half a million times more efficient than the Blue Gene doing 2000 MFLOPS per watt.
I'm just saying, the theoretical limits are a ways off yet. In the short term power costs for growing clusters for growing computing demand looks bleak maybe, but I think there's lots of elbow room yet for clever improvements. Like chilling my mineral oil in the fridge overnight before filling my desktop case with it :/ Peter On Mon, Apr 8, 2013 at 5:13 AM, Hearns, John <john.hea...@mclaren.com>wrote: > On Mon, Apr 08, 2013 at 08:58:19AM +0000, Hearns, John wrote: > > > a) humankind will inevitably demand, build and find uses for computing > systems with ever increasing numbers of fixed point of > > floating point operations per second. > > Devil's advocate time - we can simply declare that no new faster system > will be built > > Photovoltaics actually shaves off peak quite nicely, and reduces > the price of electricity. If you don't want to invest into MWh > storage (redox flow or otherwise) you could checkpoint a bit > before sunset, and resume when power is available. > > > You mean.... supercomputing types coming out in the DAYLIGHT? Working > normal hours? > That will never work. > > > Joking aside, that is a good idea. Just have to manage user expectations > when there is a cloudy day! > > > The contents of this email are confidential and for the exclusive use of > the intended recipient. If you receive this email in error you should not > copy it, retransmit it, use it or disclose its contents but should return > it to the sender immediately and delete your copy. > _______________________________________________ > Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing > To change your subscription (digest mode or unsubscribe) visit > http://www.beowulf.org/mailman/listinfo/beowulf >
_______________________________________________ Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
