[extropy-chat] ECON: Chip market responds to high oil prices...

[Dan Clemmensen]

Mike Linksvayer wrote:

>Where do you get the idea that the chipmakers are responding to
>high oil prices?  Performance per watt has been the buzz for a few
>years at least due to cooling and battery life problems.  Intel
>just got around to a complete product line revamp around this now.
>If they had been responding to recently higher oil prices I expect
>it would've taken them longer, as they would've started later.
>Besides, oil accounts for only 3% of US electricity generation:
>http://www.eia.doe.gov/cneaf/electricity/epa/figes2.html
>
>  
>
Yes. The industry has been trying to optimize efficiency since long 
before the current "oil crisis." The drivers are mobility aapplications 
on the one hand, and massive server farms on the other. From my 
perspective the server issue has been the major driver.

The fundamental problem for a massive server farm is density. Density is 
measure in square feet of floor space. This includes floor space for the 
server racks, for the cooling system, and for the backup power system.

As the power per rack goes up, the complexity of delivering power and 
(especially) cooling to that rack go up non-linearly. There is a point 
(about 10KW/rack) at which things get crazy. From a practical 
perspective, you begin to lose computing density when you cross that 
threshold. Therefore, you increase your true density (cycles per square 
foot) by using servers that need less power.

The threshold varies depending on the rental cost per square foot, but 
this cost is remarkably constant for server farms that need high-speed 
connectivity to the Internet,