OpenMP has been successfully used to parallelize large-scale application codes such as the simulation of heat transport in a rocket combustion chamber. The recent support for advanced OpenMP features such as nested parallelization also has contributed to improved scaling behavior, an example being critical point computation in a flow field  for virtual reality environments. Running hybrid parallel (MPI+OpenMP) programs on a large SMP machine offers additional opportunities to improve load balancing by dynamic thread balancing, for example in simulation of the reentry of a space glider.

Tightly integrated SMP systems such as Sun Microsystems Sunfire T2000 which employs the "Niagara" chip (officially named the UltraSPARC T1) show the way the commodity SMP node is evolving. In the usual pizzabox form factor this system supports 32 threads (8 cores with 4 threads each) with substantial bandwidth.  We will report on first benchmarks with the Sunfire T2000 as compared to a Sun Fire E2900 (a cluster of dual-core UltraSPARC IVs). Our conclusion is that such tightly integrated SMP systems are very attractive as building blocks of petascale systems.