Originally Posted by DrCloud,May 9 2008, 09:58 PM

^ Climate "modeling" (which was what we called in its early stages) was always at the forefront of supercomputer usage, along with the nukes at the weapons labs and a few other esoteric things. Now, it's "simulation" (because there are fewer compromises and more computer power), and it's still at the forefront.

I haven't kept up with what's required to do the runs these days, but they've begun using ensemble simulations not only of one model run several times (with varying initial conditions) but of different models using the same forcing scenarios, all at much higher resolution. The newer computers allow this. NCAR in Boulder just installed a 75 teraflop/s machine that will be for climate almost all of the time, but it's no where near #1 (which, as of last November, is the IBM Blue Gene system at Livermore, 478 T-flop/s or so).

The petaflop/s machine that's going in at NCSA (which might make the June list) will undoubtedly be used for a variety of things, including climate simulations -- as big computing has matured, more applications have arisen. But climate is still a big consumer. HPH

[teraflop/s = Trillion Floating Point Operations per Second;
petaflop/s = 1000 teraflop/s -- this is the next big threshold and will make headlines. A high-end home system can make perhaps a gigaflop/s, or a billion.]