Ten years ago, a 1 Tflop/s system would have put you in the top 5 of the TOP500 list, would have required a substantial data center, and government funding (from a big government). Now, you can buy a 1 Tflop/s system on the web, deploy it under your desk, and pay for it via a modest research grant. On the other hand, a top 5 system still requires a substantial data center, and considerable attention to power and cooling. But we will see the first PetaFlop/s system fairly soon. And that 1 Tflop/s system won't even make the list at #500.

These factors have enabled an explosion of the ability to use computational science, both by broadening the use of computational science via smaller, easier to use and less expensive to buy systems, and increasing the complexity of problems that can be tackled, via systems that scale higher and higher.

But not without significant challenges. The emergence of clusters based on industry standard processors and Linux has made HPC affordable, but not necessarily easier. Limitations in processor technology require trade-offs between power consumption and clock speed, resulting in the trend toward blade systems and multi-core processors, and the re-emergence of co-processors. All the while, applications performance that enables that enables computation performance requires new ways of looking at systems design, software development, data center design and even new approaches to procuring computing resources from the "cloud".

We'll talk about the technologies that enable computational science, where they are now, and where we're going.