The UK is buying a supercomputer architecture—the same type used in the world’s two fastest machines, SUMMIT and SIERRA—that is not yet widely available in the country. This matters because the UK’s current computing infrastructure lacks the memory-coherent link between graphics processing units (GPUs) and central processing units (CPUs) that this system provides, which allows researchers to tackle larger datasets and run simulations faster. The new machine, called NICE19, will primarily serve experimental scientists who generate huge data volumes—for example, from cryo-electron microscopy facilities—and modellers who use machine learning. Eight research software engineers will help these communities port and optimise their code for the new platform. If successful, the facility will reduce the “time to science” for hard problems, improve the value extracted from existing experiments and datasets, and give many UK researchers their first access to high-performance computing. The architectural similarity to the world’s top supercomputers also provides a practical route toward exascale computing, meaning this investment could shape national computing capability for years.
View original technical description
We propose to establish a new national Tier-2 HPC service ("NICE19") by purchasing and operating a novel architecture supercomputer, based upon IBM Power9 CPU and Nvidia Volta GPUs. This architecture is used in both #1 and #2 of the most recent list of the Top500 supercomputers in the world, within the USA government lab-based SUMMIT and SIERRA supercomputers. Currently this architecture is not widely available in the UK, so by establishing a national Tier-2 facility we will add significant diversity and capability to the UK e-infrastructure. 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 tackled beyond that of other GPU-accelerated architectures, increasing data sizes for accelerated simulation and analysis codes, and reducing the 'time to science' for a range of 'hard' problems. The purpose of this facility is to enable new science across many different disciplines. Our primary focus will be on experimental users who generate large data sets that need analysis (e.g. cyro-EM facilities) and for modellers who use machine learning. By supporting and bringing together these two communities, there will be many opportunities for new and exciting science. This will bring HPC to many communities who have not engage with HPC before now, with many benefits in getting better value out of experiments and existing datasets. In addition, existing investments in network infrastructure for the DiRAC consortium of STFC users will be leveraged to connect the facility to experimental facilities and other Tier-2 centres for optimal data flows between sites. This proposal is led by the N8 Centre of Excellence in Computationally Intensive Research (N8-CIR) and supported by the N8 partnership. This will provide 8 FTE of Research Software Engineers (RSEs) to support these communities, port and optimize their software for this new platform and train users, which will be essential to maximize the benefit of this novel facility. These codes will then be able to run efficiently on the facility, supported by trained users. The architectural similarity with SUMMIT and SIERRA will also provide a route to exascale computing. This proposal will therefore have a long-lasting impact, beyond the lifetime fo this facility.
Alan Real (Co-Investigator)Colin Bain (Principal Investigator)Mark Wilkinson (Co-Investigator)Matthew Probert (Co-Investigator)Paul Calleja (Co-Investigator)Stewart Clark (Co-Investigator)
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know