Dell, Nvidia power new "cloud-native supercomputer" in the UK

The expanded system at the University of Cambridge will deliver multi-tenant high performance computing for research spanning astrophysics, nuclear fusion power generation and clinical medicine applications.

Researchers at the University of Cambridge in the UK are expanding their HPC systems with servers from Dell and resources from Nvidia. Delivering multi-tenant, bare-metal high performance computing, the systems amount to "cloud-native supercomputers," Nvidia says. 

The university is adding hundreds of new Dell EMC PowerEdge servers to the Cambridge Service for Data Driven Discovery (CSD3) system, which is a UK National Research Cloud. Specifically, it's deploying more than 400 PowerEdge C6520 servers with the newly-announced 3rd Gen Intel Xeon Scalable processors. 

It's also adding more than 80 PowerEdge XE8545 servers with 3rd Gen AMD Epyc processors, as well as Nvidia A100 Tensor Core GPUs with NVLink. These systems are also using Nvidia InfiniBand networking and Nvidia BlueField-2 DPUs to offload infrastructure management, while isolating and accelerating workloads. 

Meanwhile, the University of Cambridge, along with StackHPC and funding from the DiRAC HPC Facility and the IRIS Facility, developed a cloud HPC software stack called Scientific OpenStack.

All told, the CSD3 system is expected to deliver four petaflops of application performance, enabling research in a range of areas, including astrophysics, nuclear fusion power generation and clinical medicine applications.

Another UK research institution, Durham University's Institute for Computational Cosmology (ICC), is slated to fully deploy the COSMA8 supercomputer this October. The COSMA8 will launch with more than 90 Dell EMC PowerEdge C6525 servers.

The system uses 2nd and 3rd Gen AMD Epyc processors, direct liquid cooling and Nvidia HDR InfiniBand networking. It powers research into subjects like dark energy, black holes and how the universe was formed. 

The university plans to expand the system to more than 600 compute nodes over the next year.