NICE (Northern Intensive Computing Environment) will be housed at Durham University and co-located with the existing STFC DiRAC Memory Intensive National Supercomputing Facility.
The machine consists of 32 IBM Power 9 dual-CPU nodes, each with 4 NVIDIA V100 GPUs and high performance interconnect. This is the same architecture as the US government’s SUMMIT and SIERRA supercomputers which occupied the top two places in a recently published list of the world’s fastest supercomputers.
An additional 6 nodes will use Nvidia Tesla T4 Tensor Core GPUs, targeted towards AI inference, or FPGAs. The latter will allow researchers to explore this alternative accelerator technology. This architecture supports memory coherence between the GPU and CPU and a hierarchy of interconnects to allow effective distributed GPU use, extending problem sizes that can be tackled beyond that of other GPU-accelerated architectures.
Machine usage will be shared amongst the N8 universities, but with 36% reserved for national EPSRC use by researchers from across the UK. N8 researchers will be able to apply for access to the machine at any time.
The machine will be of particular interest to the Machine Learning and simulation communities with IBM having done a lot of work ensuring common ML tools can take advantage of the architecture. Hence the main initial development work by N8 CIR will be to help users with distributed deep learning. Across the N8 there are many existing research groups that use ML who are limited in problem size that can be tackled, but with this facility, there will be a step-change in the size of problem that can be tackled.
A wide range of applications will be available on the machine, especially those already optimised on SUMMIT, including Code-Saturne, GROMACS, NAMD, AMBER, CASTEP all of which are currently used by UoM researchers on the CSF. There will also be dedicated support available in Research IT to ensure that UoM researchers make the most of this new computational facility.
Dr Simon Hood, Head of Research Platforms, UoM and Assistant N8 CIR Technical Director said “We have a growing number of GPUs on our local computational platforms, but there are never enough! The new N8 facility will give the UoM research community access not only to more GPUs, but to an architecture with memory coherence, so much larger problems can be addressed. This is great news.”