Partnership for the future – Community

Germany and Bavaria are entering an exascale future. The Federal Ministry of Education and Research and the Bavarian Ministry of Science and Art are financing the construction of an exascale-generating supercomputer at the Leibniz data center in Garching for EUR 125 million each. These systems handle at least a trillion floating point calculations or flops per second. The Garching High-Performance-System (HPC) – a novelty in procurement – is being developed for the first time as part of an innovation partnership and thus in cooperation with technology suppliers.

Innovation comes from competition – the most important basic idea in an innovation partnership for a new purchasing method. Photo: Nejk Soklic / Unsplash

It should contain state-of-the-art technology, components that are not (yet) sold and that offer new research opportunities: When purchasing its next supercomputer, the Leibniz-Rechenzentrum (LRZ) relies on a new process, an innovation partnership. The architecture of the new high-performance computer at LRZ will be developed together with technology companies by 2024. Prototypes will be built or existing components will be tailored to meet specific needs. “From 2010, it was clear that the next generation of HPC systems, exascale computers, would be developed in co-design, ie hardware and software would be optimized in coordination for the desired application. In the scientific community, many national laboratories in the US are already working on the joint development of hardware and software, ‘explains Prof. dr hab. Dieter Kranzlmüller, head of the LRZ. “We wanted to improve the ordering of our systems and an innovation partnership was the obvious choice.” . Thus, the next national supercomputer for Garching can be planned as the first in Europe under the innovation partnership. “To meet our needs,” Kranzlmüller expects, “we can research and evaluate innovative calculation methods together with manufacturers and then implement them in the new exascale system.”

A tool for modernization and economic development at the same time

Innovation partnerships have been possible in Germany since 2016. Federal and state governments support this procurement method because it supports the development of new offerings and innovative companies, and even helps modernize administrations and public institutions, according to the Federal Ministry of Economy. “The public sector spends around EUR 500 billion every year in Germany,” says Matthias Berg, head of the competence center for innovative procurement at KOINNO. As a consultant, he has already accompanied many such partnerships. “If only one percent affects new products and services, that is a huge boost to the economy.” The development of the exascale computer is not so much about modernization and economic promotion as it is about the development of promising computer technologies. Like other Innovation Partnerships, LRZ will also operate in several phases. About ten specialists from various fields are involved in the procurement, which started at the beginning of 2022 with the competition. In several rounds and until mid-April, hardware manufacturers competed according to LRZ specifications with the concepts of another high-performance computing system or HPC system. Bidding companies are now invited to submit official LRZ bids along with detailed co-design steps. The agreement should enter into force in July and the companies will start practical development work on the new system. By 2024, it will be clear which company will build and support the next exascale-generation supercomputer.

“Our innovation partnership enables us to develop and optimize the technology for the next system with several vendors at the same time,” says Herbert Huber, citing the advantages. The PhD in Physics has been head of the high-performance systems division at LRZ for ten years and has already assisted with the planning and procurement of several supercomputers. “That way, we can get to know companies and the way they work long before we deliver a high-performance computer.” The Big Data & Artificial Intelligence Team (BDAI) and the Computational X Support (CXS) group, which supports scientists in optimizing algorithms and implementing them in HPC systems, are also working on proposals and specifications. We sincerely hope that the interaction of the supercomputer components can be tested and improved beforehand. In addition, the startup difficulties should decrease once the system is booted. “After explaining the first technical steps and contracts,” Huber plans, “we will also involve users in the development work.”

Effort, discussions and risk

Innovation partnerships are more complex than other purchasing processes: “They pay off for high-investment and new solution products,” notes Berg. “However, there is usually only a vague idea of ​​the technology or service to be developed, and contracts for research and development as well as prototype building are made.” They touch upon copyright, patent and marketing rights and are therefore drawn more individually and in more detail. In addition, many precautions must be taken. Among other things, LRZ must ensure that all companies can work with the same information and that no one learns about the technical innovations of others. So, prototypes of processors, accelerators and other components are installed in different and closed racks. And the team swears to remain silent about who contributes what and how to the next supercomputer is currently not allowed to be made public.

Innovation expert Berg also recommends not to underestimate the time required. “For each step in the development, the design specifications, and most importantly the test criteria, need to be precisely defined and developed,” he says. “As many people as possible should be involved in the discussions to take account of different perspectives and needs.” Even if several companies are working on a problem in the first round and are rewarded for it, Innovation Partnerships also help save money. Experience has shown that costs in the IT sector are cut by up to 30 percent as hardware or software is developed more specifically with the real needs in mind. Above all, however, for the new Garching supercomputer, the best possible exascale system for research and science should arise through technological competition – and hopefully at the same time, new components will also be created that will help better design IT processes in the economy. In an innovation partnership, participating companies run the risk that their solutions will not ultimately be selected and structured. However, they can independently market the products created during the cooperation and adapt them to other possible applications. “It’s quite possible that ideas will fail, but the experiences are valuable to everyone involved. Failure is part of innovation, the public sector has a hard time because it has to account for its expenses, says Berg confidently: “Although the process is complex and the risks are high, I have never seen an innovation partnership fail.” (vs)

Leave a Comment