Advances in "cloud computing", clustering, and general-purpose computing with commodity GPUs suggest compute power per dollar may increase significantly faster than Moore's Law predicts.
The first full week of January is always an exciting time in the computer industry. The Consumer Electronics Show and MacWorld bring announcements of new products; computer firms that aren't oriented to the retail consumer also tend to make new announcements, in order not to be left out of the news. This week, several announcements suggest that a more distributed "cloud computing" model will drive large system architecture, providing increasing computer power to the user even more quickly than Moore's Law would suggest.
One of the keystones of cloud computing is the ability to build clusters with high-speed interconnections. While there are specialized inter-computer connections, like Infiniband, high-speed 10 Gb Ethernet is also quickly becoming practical, while not requiring as much specialized new network hardware and different management. This means that high-capacity clusters can be built at lower cost.
At the same time, AMD has announced a new planned supercomputer based on general purpose computing with commodity graphics processors. This has many potential uses, but AMD in particular is looking at their new machine to provide high-performance computing directly to the consumer for gaming and graphics.
Architectures based on clustering using 10Gb Ethernet, powered by cheap supercomputers built on commodity graphics processors, combine to create an environment for high-capacity inexpensive computing to the consumer. These architectures have dramatic implications for gaming, home video, and network-based computing.