The University of Bath has launched a new, cloud-based supercomputer, Nimbus, which the university is calling the “central pillar” of its new portfolio of cloud research computing environments. The university says that the launch of Nimbus marks it as the first university in the UK to move all of its high-performance computing workloads to cloud environments.
Nimbus, which is hosted on Microsoft Azure, is a bit amorphous — the university does not provide a number of nodes, instead saying that the system offers “virtually unlimited on-demand HPC.” They do, however, specify the different configurations available on Nimbus via Azure: the HB, HBv2 and HBv3 instances, offering 16 to 120 AMD CPUs; the HC instances, offering 44 Intel CPUs; the Fsv2 instances, offering 2 to 72 Intel CPUs; and the NCv3 and NDv2 instances, offering 1 to 8 Nvidia Tesla V100 GPUs. The nodes are networked with Mellanox EDR InfiniBand (100Gb/s or 200Gb/s) and supported by a little more than 256GB of data storage (alongside “full integration” with the University of Bath’s home drive).
“Our new Nimbus research computing infrastructure opens up many new possibilities for innovation and represents a major step forward for our research capabilities,” said Ian White, vice chancellor and president of the University of Bath. “This access to huge computational power will be an important tool in our armory as we seek to achieve our ambitious research goals and tackle major societal challenges in areas such as energy, transport, public health and sustainable living.”
Nimbus is supported by a series of other HPC resources, predominantly cloud-based. The university controls 15% of the compute time on the Cray-built, Arm-based Isambard system, a collaboration between Bath, its three sister universities, and the Met Office. An Azure-based teaching cluster offers HBv2 instances with AMD CPUs and “up to 4 teraflops of FP64 compute” and is complemented by several University of Bath courses that have now migrated to teaching on cloud resources. Meanwhile, an on-site HTC cluster, Anatra, equipped with eight compute nodes, each with dual AMD Milan CPUs and 256GB of memory — Bath says that this is for “software applications that can’t run in the cloud.”
Nimbus — which was announced last year under the name “Janus” — is set to succeed the on-site Balena system, which is equipped with a couple hundred Intel-based nodes (Ivy Bridge and Skylake) and a smattering of AMD, Intel and Nvidia accelerators.
Explaining the move to the cloud, the university cited a warning from the Alan Turing Institute that the Tier 1, 2 and 3 systems in the UK “are either working at maximum capacity, under-resourced, limited in their compute provision specifically to support AI, or a combination of all three.”
Nimbus was launched at an event late last month that included speakers from Microsoft and the Met Office. The Met Office, of course, is in the midst of deploying its own Microsoft Azure-based weather and climate supercomputer.