This morning Southern Man was treated to a pair of lectures by the always dynamic and entertaining HN...
HN resolves a minor technical glitch.
...on the future development of supercomputing and on scientific libraries for high-performance computing. He'd heard the second talk before but always learns new things anyway. A few fun facts:
- Ten years ago the university supercomputer topped one TF and cost just under a million dollars. Today you can buy a graphics card with TF performance for a few thousand dollars. That's ten years from supercomputer to desktop.
- Moore was righter than he ever realized; speed increases in everything are exponential. But the slopes on all the lines are different; some things get faster faster than other things.
- Several years ago the emphasis was on clock speed. A few years ago the emphasis was on number of cores. Today the bottleneck is communication and the emphasis is on finding ways to get data to the cores.
- As an example of the previous point, the fastest math packages are optimized for cache performance, not for supercomputing performance.
This cool "fort" was near the Duck Pond.
Afternoon was free time so Southern Man knocked out several more Euler Project problems and has re-learned some C++ that he had forgotten. He's looking for problems his beginning programmers can do as exercises as well as problems that can be easily adapted to multicore algorithms. And as there was no open lab after dinner he retired to his nice cool hotel room for movies and relaxation. Tomorrow is the final session!