End of 2010, the Center for Computing and Communication at RWTH Aachen University (RZ) signed a contract with Bull to install a new 300 Teraflops HPC system until May 2011. Please take a look here >>> (German only) for more information. The system was handed over to the RZ in June 2011.

The cluster will consist of two main partitions, the MPI partition with 1350 2-socket nodes with Intel Westmere EP processors and the SMP partition with 358 4-socket modules with Intel Nehalem EX processors.

Either two of four of these 4-socket modules can be coupled by propriatary BCS chips to build  8-  or 16-socket systems, respectively. The RZ has come to an agreement with the company Bull, to change the original plan to install these modules as 8-socket nodes, but to start with 4 socket nodes initially and then later , in October 2011, to couple four of these modules and operate those as compute nodes with 16 sockets.

These very large compute nodes with 16 8-core processors each will provide up to 1 TeraByte of main memory providing an opportunity for a new class of applications.

A few of these 16-socket systems will be provided to the RZ early on, but as the optimization processos is still ongoing, access to these system will be limited and confidential.

Furthermore 16 of these 4-socket modules will be coupled by the vSMP software of the company ScaleMP to generate a singles system with 512 processor cores accessing 4 TeraByte of shared memory (minus overhead). The efficient usage of such a system requires a sensible way of NUMA-aware programming.

The tables give a summary of the configuration of phase 1 (May 2011) and phase 2 (October 2011).