Usually start with a few small nucleation sites. Create enough of a net savings in energy, and entropy, and whammo, you are starting the rapid, highly nonlinear, often discontinuous traversal of the phase coordinate.
Of such things, revolutions are born within computing. It is happening with APUs, it has happened with dual core, and it appears more likely to be happening outside of HPC.
Update: R.L. Polk is not that far from us. Seems to have taken them a long time to do migration, and I am concerned that someone was not reporting everything, or analyzing it correctly. That said, a story this morning points out what I have been saying for a while: deploying Linux lowers costs, and allows you to save time. This is true whether you have a loosely coupled cluster (e.g. grid), or a tightly coupled cluster. R.L. Polk will not be paying per connected client, they need no CALs, so as their user base scales, their costs do not need to scale, until they decide to add more processing power into their grid.
All businesses like to lower costs. All business like to do more with less.
If you are deploying a 10-1000 node cluster or a 100 node grid, the infrastructure cost, the recurring costs, and the acquisition cost are all noticeable. The higher they are, the more value they need to bring. If you do a cost benefit analysis between two roughly equal choices, one costing large acquisition and user access fees as compared to the other, the lower cost item will usually win out unless there are other considerations. The TCO gets biased even more in the favor of one of the solutions due to security and exploitation risk issues.
This is one of the reasons why HPC has been so explosive in growth. The TCO and upfront costs as compared to the alternatives are much lower. The RLPolk work is not HPC (any more than eBay is). But they benefit from the same issues that HPC does. The larger market does as well.
I expect to see this in desktop/laptop systems soon as well. I run my laptop as a dual boot Linux and Windows XP system. I don’t have any plans of purposefully loading Vista on it. I don’t need it, and from everything I am hearing about Vista DRM and bits, I simply don’t see any positive value in it. XP is fine for windows specific tasks at the moment. Heck, W2k was fine, I didn’t need to go to XP. Linux OTOH allows me to compute securely. If I destroy something, it is confined to my personal stuff, not the machine: its that principle of not running as an administrative user. Everything works on that laptop: Graphics, USB, Firewire, bluetooth, network, wireless, under Linux. I am not as vulnerable to keyloggers and malware installing themselves as I am under XP.
As stated in the title, it is only a matter of time. The CBA is strongly tilted against windows at this point, as is the risk profile.