The recent installation of 14 new GPU nodes on the Raijin supercomputer will offer significant performance increases for users running GPU-optimised code.
Chemistry, molecular dynamics and bioinformatics projects who have already transitioned to the GPU's have experienced faster results and increased efficiency.
This upgrade gives NCI an additional 163 Tflops, an overall 14% performance boost.
Each compute node includes four NVIDIA Tesla K80 GPU Accelerators, supplemented by two sockets of 12-core Intel Haswell E5-2670v3 running at 2.3GHz and 256GB DDR4 system memory. In addition, each K80 houses 24GB of GDDR5 memory, bringing the total memory of each GPU node to 352GB.
All of NCI's GPU nodes combined give a total of 279 552 NVIDIA cores available for use. The addition of these GPU nodes will deliver tangible increases in computational speeds to researchers and other users that require exceptionally fast turnaround times.
A small collection of GPU-ready programs have already been installed on Raijin /apps including:
- Chemistry: NAMD, GROMACS, AMBER, LAMMPS, VASP
- Bioinformatics: nvbio(nvBowtie), nextgenmap
- Deep Learning: tensorflow, theano
- Other: cuda, openmpi, Matlab
Please visit https://broken.link/gpu for more information on the system configuration, how to submit a gpu job and some of our GPU benchmark results.
The purchase of the nodes was made possible through NCRIS 2015 funding.
