High-Resolution Weather Forecasting: A Teraflop Sustained on RISC/cache or Vector Processors
Stephen James Thomas, Michel Desgagné, Michel Valin · Kluwer Academic Publishers eBooks · 2005
The next generation of supercomputers will be based on clusters of symmetric multi-processor (SMP) nodes interconnected by low-latency, high-bandwidth networks. Individual nodes will contain from 1 to 128 RISC/cache or vector processors in a shared or distributed-shared memory (DSM) configuration. These machines support a hybrid programming model combining sharedmemory threads (Open/MP) with distributed-memory message-passing (MPI). We discuss the various design considerations, trade-offs and constraints in the construction of high-performance atmospheric models capable of achieving one TeraFlop of sustained performance. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.