Momentum for Accelerators is accelerating! Green and lean, for some very critical HPC workloads, they can deliver +10x-20x the performance per price, per watt, per foot of standard servers. HP, working with its partners Nvidia, Voltaire, and NEC, has just been awarded the 2.4 PetaFLOPS TSUBAME2.0 system by the Tokyo Institute of Technology (TiTech). This system has approximately 17,000 cores in a mix of Intel Westmere and Nehalem processors in HP ProLiant GPU-optimized servers and approximately 4,200 NVIDIA Tesla GPGPUs, and will use a Voltaire-based InfiniBand fabric. For software, the system will offer cloud services for its research and academic users, and provide a choice of Linux and Windows HPC Server. HP Cluster Management Utility (CMU) will be deployed to manage the cluster, leveraging new CMU features to simplify managing and monitoring GPGPUs. More information, for those of you who can read Japanese (or have a good web translator tool), see TITECH: http://www.gsic.titech.ac.jp/sites/default/files/p
With an evolving ecosystem to ease development, deployment and management, GPU-optimized servers will move into the mainstream HPC community, as users seek to leverage the same capabilities deployed at TiTech, Georgia Tech, and key industrial users (such as oil and gas). HP is working to enable that evolution, with new system designs, comprehensive support and seamless integration into standard cluster offerings.