NVIDIA, IBM and Mellanox working together HPC

NVIDIA, IBM and Mellanox working together HPC
NVIDIA today has evolved into a company that not only produces GPU for gaming cards. GPUs are increasingly used in science and data centers where it is possible to divide the parallel computing among the thousands of stream processors. NVIDIA receives a significant part of the turnover in this field in the future, it is likely to grow.

The close cooperation between NVIDIA and IBM for anybody not a secret, in the future, the GPU “Pascal” in IBM servers for the first time will use NVLink. Now, the company announced the creation of the center of Montpellier in the French “POWER Acceleration und Design Center”. It involves not only the creation of NVIDIA and IBM, but a specialist in the field of high-speed networks and Mellanox manufacturer of motherboards Tyan.

These three companies thus attempt to further promote his initiative OpenPower Foundation. All partners are added to her own technology: the architecture of IBM POWER, GPU NVIDIA accelerators and in general, the interconnection NVLink, and Mellanox has considerable experience in creating high-speed network (InfiniBand). In the center POWER Acceleration and Design Center partners will be able to optimize and test their hardware and software solutions in the context of OpenPOWER Foundation. Also planned a variety of projects in the field of High Performance Computing.

The first fruit of the collaboration of three companies will be two supercomputer “Sierra” and “Summit”, which in 2017 should take the lead among supercomputers. The customer is the US Department of Energy. “Summit” will be used at the Oak Ridge National Laboratory, and the “Sierra” – at the Lawrence Livermore National Laboratory. Lawrence (LLNL). “Summit” will reach capacity of between 150 and 300 petaflop and “Sierra” will replace the IBM BlueGene / Q in Livermore with a capacity of 100 petaflop. For comparison, the supercomputer Tianhe-2, the current leader of the list of supercomputers, gives only 33.8 petaflop.

In the Jülich Research Center, IBM and NVIDIA also plan to create a similar design center, but on a smaller scale than in Montpellier .

NVIDIA, IBM and Mellanox working together HPC updated: July 7, 2015 author: Jonathan Davis