Skip to main content

Public Procurement of Innovative Solutions for High-Performance Computing

Periodic Reporting for period 3 - PPI4HPC (Public Procurement of Innovative Solutions for High-Performance Computing)

Reporting period: 2020-04-01 to 2021-09-30

In this project, a group of leading European supercomputing centres formed a buyers group to execute the first joint Public Procurement of Innovative Solutions (PPI) in the area of high-performance computing (HPC). With the additional co-funding by the EC, the total investment was about € 73 million and a significant enhancement of the HPC infrastructure could be realised. The involved HPC centres, namely BSC, CEA/GENCI, CINECA and the Jülich Supercomputing Centre (JSC), have a strong track record in providing supercomputing resources at the European level.
This joint initiative was designed to create benefits in multiple respects:
▪ More advanced supercomputing resources are now available for science and engineering applications in Europe through PRACE, the pan-European HPC infrastructure, thanks to the new supercomputers Joliot-Curie at CEA, Marconi100 at CINECA and JURECA-DC at JSC. Furthermore, a large-scale high-performance storage infrastructure has been realised at BSC.
▪ R&D on HPC architectures and technologies in Europe have been strengthened as strong preference was given to solutions that integrate innovative solutions where many have been developed in Europe. This includes, e.g. solutions for monitoring and managing power consumption to improve energy efficiency.
▪ Finally, the project was expected to generate more impact on the design of the solutions according to the needs of scientists and engineers in Europe thanks to the coordinated approach. Various success stories show that the deployed solutions are indeed suitable for enabling new research in Europe.
The participants worked together on coordinated roadmaps and based on the outcomes of a joint market consultation, common requirements and evaluation criteria were jointly formulated. These were complemented by site-specific requirements and evaluation criteria. Final procurement contracts were designed to be awarded within four different lots. By deploying a variety of innovative solutions, a broad range of capabilities is now available.
During the initial phase, the project focused on analysing common needs and identifying common technical goals and requirements. A framework for executing a joint procurement was developed. The results of this initial planning were used to inform the market and enter into a market consultation phase. Feedback from the market has been taken into account in the final tender documents that were published in May 2018.
The joint tender was organised in four lots, one for each of the involved HPC sites. For each of the lots, up to five suppliers requested participation in the competitive dialogue. The qualified candidates were jointly selected. Afterwards, each of the lots continued with a competitive dialogue at the local level. All local tenders were completed successfully and the last contract was awarded in early 2020.
BSC deployed a new high-performance storage infrastructure that can scale to more than 100 PByte as a single file space or file system. CINECA deployed a supercomputer optimised for compute-intensive applications with a peak performance of 32 PFlop/s. CEA/GENCI deployed a versatile supercomputer with a peak performance of 22 Pflop/s supporting highly scalable computing applications as well as high-performance data analytics and machine learning applications. JUELICH deployed a supercomputer with a peak performance of 16 PFlop/s that is optimised for data-intensive applications and therefore features a large memory and storage capacity.
During the final phase, the project focussed on the evaluation of the innovative solutions that had been procured. The benefits of the most important innovations were assessed, which included solutions for integration of processors with a particular large power envelope, liquid-cooled power supplies, novel network topologies, integration of fast storage devices and novel hierarchical data archives. For all procured compute systems a significant improvement of the power efficiency could be confirmed.
The PPI4HPC project was the first project to perform a joint procurement of innovative HPC solutions involving public procurers from four different EU member states. The buyers’ group jointly identified a set of common challenges towards future exascale capabilities, which will require a significant increase of the system’s performance, while keeping power consumption constant or allowing only for moderate increases. A further common challenge was to improve on versatility to serve a broader range of applications, in particular applications employing machine learning techniques. To strengthen the focus on deploying innovations, a mandatory innovation criterion was added to the tender, which forced all suppliers to explicitly identify the innovative features of the offered solutions.
All deployed solutions did improve significantly concerning the state-of-the-art as of 2018:
* New solutions for HPC system integration facilitate the integration of CPUs with much higher power envelopes and the use of liquid-cooled power supplies.
* Novel network topologies opened the door to future exascale systems with a drastically larger number of nodes.
* Storage solutions that scale in terms of capacity to new levels.
* Drastic improvement of power efficiency for all deployed HPC systems.
The joint procurement executed by PPI4HPC had the most important impact in the following areas:
* New science results, in particular in areas related to grand challenges like climate research, energy research, new materials and health with the upgrade (about 70 PFlop/s) of the European distributed HPC infrastructure (PRACE) according to the needs of scientists and engineers.
* Improved energy efficiency and therefore energy savings as all procured solutions implement innovation in this area.
* Path towards more European procurement of high-performance computing and storage solutions, increasing the willingness of multiple publicly funded entities to work together on some form of joint and coordinated procurements in the area of HPC.
Today's flagship computers at the PPI4HPC supercomputing centres