The Swiss National Supercomputing Center, or CSCS, teamed up with Nvidia and HPE to launch a new supercomputer named Alps, which went live at the end of 2024. Alps is now the seventh most powerful supercomputer globally and the second most powerful in Europe. I had the opportunity to chat with Thomas Schulthess, CSCS’s director and a professor at ETH Zurich, to get the scoop on Alps.
First off, Schulthess explained that CSCS is more than just a supercomputer; it’s a hub of people and expertise located in Lugano, Switzerland, since 1991. While many organizations run their own weather models, CSCS operates the supercomputers for MeteoSwiss, making their partnership particularly strong over the years.
Alps is not just another supercomputer but a unified platform that aims to tackle the rising challenges in data and complex workflows in scientific computing. The journey to Alps started back in 2015 after a peer review suggested they needed a new approach following the success of Piz Daint, Europe’s first GPU-based supercomputer, which had been deployed around 2012-2013.
The design journey wasn’t straightforward. Initially, they partnered with Cray, now part of HPE, which was pushing towards a micro-service or cloud-native architecture. Despite facing significant hurdles along the way, they moved forward with the procurement in 2019 and ultimately opted for a mix of computing architectures from both Nvidia and AMD. So now, Alps features Grace Hopper processors from Nvidia along with AMD’s MI300A accelerators.
When you look at how Alps operates today, it’s quite impressive. It utilizes a vast slingshot network, akin to Frontier and LUMI, where you can partition it into storage devices or compute nodes. It supports multiple architectures like Grace Hopper and AMD MI300A, as well as older AMD Rome-based nodes, making it versatile for various workloads. The focus is on fostering collaboration with software developers through robust application support.
Schulthess likens the user experience on Alps to a cloud service. They provide Infrastructure as a Service (IaaS) to research institutions, such as the Paul Scherrer Institute, which dives into large user programs. Other research areas, including AI and climate modeling, also benefit from dedicated platforms. Recently, MeteoSwiss’s numerical forecasting system ICON transitioned to run on Alps, highlighting the platform’s capacity and flexibility.
Funding for CSCS and Alps comes primarily from the ETH Domain, with additional resources from MeteoSwiss and third-party collaborations covering about 20% of the costs. Being publicly funded means they can’t engage in commercial activities but are open to research partnerships where costs are covered by the collaborating entities.
Lastly, Schulthess is involved in the OpenCHAMI consortium, which includes notable institutions like Los Alamos and Lawrence Berkeley laboratories. This consortium is working on developing future system management infrastructure, with Alps playing a crucial role in this evolution over the next few years.