NCI has augmented its growing segment of GPU infrastructure with 8 new Pascal P100 GPU accelerators from NVIDIA. These new accelerators represent the next-generation of graphics processors for supercomputing. As evidenced by the rapid uptake and continuing demand for the 120 Kepler K80 GPUs installed in the past year, there is clear interest in the Australian computational research community for GPUs as part of future HPC infrastructure.
Certain programs are particularly suited to the strengths of GPUs. Widely used chemistry packages such as VASP, NAMD and GROMACS have all been optimised to run on the new GPU nodes, as well as bioinformatics codes such as NVBio and deep learning codes such as TensorFlow.
High performance scientific computing is going through a paradigm shift towards the use of machine learning and artificial intelligence in research. To enable this research, NVIDIA has specifically designed the P100s with a high speed interconnect called NVlink – delivering 120GB/sec of bi-directional bandwidth between the GPUs. In addition to machine learning, many scientific applications will benefit from the new high performance interconnect. These new machines will provide extraordinary new compute capabilities to NCI researchers. The machines are now part of NCI's flagship supercomputer and have access to over 40PB of Lustre storage.
For more information about the NCI GPU systems, please visit nci.org.au/gpu.