This has always been the case, too. In school I once read somewhere that if you knew you had a computation to do that was 100 MFLOPS (or whatever seemed like a

This has always been the case, too. In school I once read somewhere that if you knew you had a computation to do that was 100 MFLOPS (or whatever seemed like a large number at the time), and you had to complete it sometime in the next 18 months, a solid strategy is to wait 9 months and buy more-modern hardware. You'd end up with more 'residual' computing power (when your computation is done, you have the hardware lying around, and now it's more powerful than if you had bought on day 0), and it would probably cost less in energy as well.