EuroHPC

The European High-Performance Computing Joint Undertaking (EuroHPC JU) is an international partnership tasked with developing a pan-European supercomputing infrastructure. Its objectives are:

  • making supercomputer resources easily accessible to domain scientists

  • assisting simulation software transition to the pre-exascale era

  • enabling the pooling of existing European HPC resources

  • building new exascale HPC facilities in Europe

The EuroHPC JU is collaborating with HPC facilities, industry partners, National Competence Centres (NCCs), the Horizon Europe funding programme which fosters a multitude of international HPC initiatives, as well as the Partnership for Advanced Computing in Europe (PRACE).

PRACE is a pan-European HPC organization for the strategic democratization of supercomputers. This is achieved through the standardization of HPC environments, the organization of recurring C++/MPI/CUDA/HPC training events, and the provision of one-click installation scripts for major compiler toolchains (GCC, Intel), libraries (OpenMPI, MPICH, CUDA, Boost, LAPACK, etc.) and simulation software (GROMACS, LAMMPS, ESPResSo, etc.).

In addition, PRACE continuously publishes Access Calls to apply for compute time. Applications are open to all academic institutions and industries located in the European Union, as well as institutions funded through a Horizon programme. The EuroHPC JU coordinates these calls and oversees the selection process to ensure a fair allocation of compute time.

Most facilities managed by the EuroHPC JU are Tier-0 supercomputers [Axner, 2024] (European centers with petaflop peak performance) and Tier-1 national centers. Between 35% and 50% of their compute time is allocated through PRACE calls.

Access modes

EuroHPC Access calls are continuously open and declined in multiple flavors:

  • Development Access: once a month

    • project can be 6 months or 1 year long, renewable up to 2 times

    • the evaluation process takes 1 month

    • the project stated goal must be about algorithm development and optimization (this includes benchmarking of different algorithms or libraries)

  • Benchmark Access: once a month

    • project can be 2 months or 3 months long, and must start within 2 weeks once approved

    • the evaluation process takes 2 weeks

    • the project will collect performance data on specific hardware in order to assess the feasibility of using that hardware in an future regular or extreme scale call

  • Regular Access Mode: thrice a year

    • project can be 1 year long, renewable

    • the evaluation process takes 2 months

    • the application must be supported by a positive-looking benchmark report

    • the project will use a petascale supercomputer to carry out large scale simulations that cannot be carried out with lower-tier clusters

  • Extreme Scale Access Mode: twice a year

    • project can be 1 year long, renewable

    • the evaluation process takes 3 months

    • the application must be supported by a positive-looking benchmark report

    • the project will use a pre-exascale supercomputer to carry out large scale simulations that cannot be carried out with petascale supercomputers and that have a potential for high international scientific impact

The specific details of each access mode can be found in the EuroHPC JU Access Policy.

Applying for compute time

Writing a EuroHPC Access proposal is time consuming. Besides documenting the algorithms and their parallel performance, one has to measure the amount of data communicated between nodes via MPI and make accurate prediction of the amount of RAM, GPU memory and disk space that will be used, and determine the corresponding bandwidth. PRACE provides data sheets for all supercomputers eligible in a call, which specify the amount of node RAM, GPU memory, disk space and bandwidth available on that hardware.

Applications are usually around 10 pages long. See Proposal ID EHPC-DEV-2023D05-050 for an example of a successful application for Development Access on Vega. Consult Selected publications to find relevant HPC literature to strengthen your application, e.g. benchmarks on clusters that detail the performance characteristics of your software or algorithms, or software engineering papers that support the extensibility and re-usability of your software.

Obligations

Accessing EuroHPC facilities comes with strings attached:

  • scientific publications and dissemination material must acknowledge the supercomputer

  • a report must be submitted at the end of the access period (within 3 months) outlining the results of the project; quoting the guidelines: “failure to submit a Final Report may disqualify future proposal submissions to EuroHPC by any member of the research group”

  • the results of a regular access or extreme scale access must be published in the scientific literature and presented at HPC conferences and events

In addition, research results carried out in partnership with industry can be advertised in the FF4EuroHPC and the EuroCC success stories booklets.

Publications

Booklets