Periodic Reporting for period 1 - ACROSS (HPC BIG DATA ARTIFICIAL INTELLIGENCE CROSS STACK PLATFORM TOWARDS EXASCALE)
Reporting period: 2021-03-01 to 2022-08-31
Overall Platform Objectives:
-Foundation and co-design of energy efficient and HPC/BD/AI cross-stack platform.
-Validation of ACROSS platform through industrial applications and fostering societal applications adoption.
-Ensuring ACROSS platform interoperability, adaptability and security.
Pilots specific Objectives:
-Aeronautics: Enhancing effectiveness in designing key aeronautical components (engine components, combustors, turbines) by adopting new workflows, Multi-scale/Multiphysics unsteady approach, and artificial intelligence.
-Weather and Climate: Enhancing global numerical weather prediction by means of hardware acceleration, low-latency exploitation of climate simulations, and enabling HPDA on large datasets.
-Energy and Carbon Sequestration: Improving capability of performing large-scale carbon geologic sequestration simulations; enable direct subsurface flow simulations on processed seismic data.
Technology-Specific Objectives:
-Co-designing HW/SW integration in collaboration with pilots, for creating exascale-ready services and ensuring compliance with the future EPI initiative.
-Building a platform supporting the execution of large-scale HPC simulations along with exploitation of big data and deep learning techniques and with seamless exploitation of hardware heterogeneity.
-Efficient execution of mixed HPC/BD/DL workflows through hardware accelerators.
In term of society impact and relevance, ACROSS project copes with the rapidly evolving HPC, Big Data and AI stacks, exploiting hardware and software architectures for complex workflow applications. The results will enable or drastically improve performance of a wide range of computationally intensive tasks, such as large-scale complex simulations and advanced AI-enabled analysis aim of generating innovation and value creation in key societal and industrial key sectors in Europe as (Greener aero engine, Weather and climate and C02 seqestration
In term of dissemination:
· In total 7 scientific publications have been reported including 1 book chapter, 2 journal papers, 2 conference papers, along with 2 scientific posters.
· Publications in wider audiences, like ETP4HPC project handbook 2022 and HiPEAC info magazine (issue 66).
· 12 blog posts already published.
· Reaching 3100 visitors in the project website, 360 social media followers.
· Participation in 25 events, conferences,
· Organisation of 2 training events, 1st tech-forum of the ACROSS project.
· Preparation and circulation of 2 newsletters.
The main exploitation results of the first held of the project comprise of:
· Identification of 19 exploitable results.
· An initial business model ready.
· A plan for the market analysis and business modelling of the project.
· Organisation of the 1st internal workshop for market analysis, called “ACROSS Strategic Planning”.
-Stack convergence between HPC, Big Data and Machine Learning.
At M18, ACROSS moved on towards its main goal of creating a convergent platform supporting HPC, BD and ML tasks. Specifically, project partners defined the HW and SW building blocks as well as deeply analyzed pilot workflows.
-Hardware accelerators.
At M18, ACROSS investigated on the best solutions in terms of programming models, performance, and productivity. Some effort has been spent in investigating the potential benefits brought by neuromorphic architectures.
-HPC orchestration.
At M18, ACROSS collected all the pilot and system requirements and transformed these into a modular multi-level orchestration architecture.
-Aeronautics manufacturing.
At M18, ACROSS started to improve over the U-Therm3D solution. On the side of the turbine design, the design approach has been based on a workflow constructing a large database of optimal geometries, by relying on HPC and HPDA solutions, as well as integrating ML/DL techniques.
-Weather, climate and farming.
Ast M18, ACROSS worked on developing and integrating innovative technologies for data-stores and semantic-aware indexing, such as storage-class memory, addressing the current I/O bottleneck and enabling the improvement of the (spatial) precision of the models.
-Energy and carbon sequestration.
At M18, two main directions of improvements over SotA have been investigated: the integration of the Damaris framework with OPM flow reservoir simulation tool, and the improvement of OPM flow itself, including the use of accelerators for reservoir simulation.