The UK's Met Office, which Microsoft helped buy a £1.2bn supercomputer, is twice as powerful as the country's current supercomputer and ranks 25th in the world. Now, HPC Wired has revealed more details about the supercomputer. It is expected to be the world's most powerful supercomputer in the weather field when it is activated in 2022.
First, the Met Office will get four HPE Cray Ex supercomputers powered by AMD Epyc Milan server processors.
In addition to integration with Microsoft Azure, this delivery includes an active data archiving system capable of supporting nearly 4 exabytes of data.
Four supercomputers provide peak computing power of more than 60 p-FLOPs, a sixfold increase over the agency's existing system.
Second, after five to eight years of operation, the system will use an upgraded fourth-generation Xiaolong processor.
AMD Genoa CPUs are expected to triple their current performance, and supercomputers are expected to be 18 times more powerful by then.
Richard Lawrence, IT Commissioner at the Met Office, explains:
We realized that the next purchase had to do something different, because it takes us on average about two years to buy any new supercomputer, and then another year to actually get it up and running.
For projects that require multiple purchases, this wasted time can be extremely valuable. So we started thinking proactively, trying to save more time on supercomputer procurement, so we could spend more time running the equipment.
In a departure from previous operations, the Met Office will not manage the supercomputers itself but will entrust them to Microsoft as a "high performance computing as a Service".
The new system has large memory capacity and enhanced nodes, complemented by a variety of storage infrastructure.
"Microsoft is going to give us a fully functional supercomputer that will take care of all the aspects of hosting and using it internally," he added.
For the upgraded second generation service, a mechanism has also been established in the procurement agreement to analyse available resources in the market and to ensure that mid-stream updates will enable the Met Office to meet its performance targets.
Interestingly, for the first time, four clusters of systems will also be hosted off-site (two for each in two separate locations), further increasing the resilience of the system.
This would also allow the Met Office to switch to a standby operating environment if it needed to "patch" a supercomputer, or if one of its systems clusters failed.
In the end, although this is the largest investment the firm has ever handled, experience suggests that most people will find it worth it.