Owen Densmore wrote:
> Seems like GPU's are not only as powerful for computation as
> mentioned on Friam in the past, but now tools are becoming available
> for programming them.
>
Here is _the_ hardware for massive agent models stalled on memory latency.
Put it on your supercomputing holiday wish list!
http://www.cray.com/products/xmt/index.htmlThe XMT is built upon the Sandia Red Storm architecture, but using
proprietary Cray Threadstorm processors which are pin-compatible with
Opterons. Auto-parallelizing compilers implementing a fancy kind of
OpenMP, and hardware blocking memory.