European Processor Initiative
European project developing low-power processors and accelerators for exascale computing.
The European Processor Initiative (EPI) was a strategic European project that created a roadmap for a new family of low-power processors and accelerators. These were designed for extreme-scale computing, high-performance computing (HPC), big data, edge computing, and emerging applications like AI and automotive. The project aimed to develop European microprocessors and accelerators to support technological autonomy and sovereignty, serving as core components for future European supercomputers to tackle societal challenges, boost innovation, and compete globally in the HPC market.
The project was split into two phases under Specific Grant Agreements. Phase 1 (SGA1) ran under the Horizon 2020 program from December 2018 to December 2021. Phase 2 (SGA2) began in January 2022 and concluded in March 2026, funded by the EuroHPC Joint Undertaking, the European Union, partner member states, and internal partners. In the second phase, the focus was to finalize the first generation of low-power microprocessor units and accelerators for European exascale machines, start developing the second generation, and ensure paths for industrialization and commercialization. The project concluded in 2026 with significant outcomes.
The consortium developed general-purpose processors (GPP) using ARM architecture and accelerators based on RISC-V ISA, targeting integration in Bull Sequana systems. In 2019, the company SiPearl was incorporated to commercialize the EPI GPP technology. Hardware design was supported by collaborative hardware/software co-design and a full-stack software ecosystem for both GPP and accelerators. Specific application domains were selected for validation: automotive in Phase 1 and AI in Phase 2.
After Phase 2 ended in 2026, SiPearl demonstrated the Rhea1 chip running a complete HPC software stack. It booted on a test board with a standard Linux distribution, high-speed HBM memories, and hardware performance monitoring counters. It executed HPL and STREAM benchmarks, showing readiness for exascale HPC workloads. Rhea1 features 80 Arm Neoverse V1 cores, over 61 billion transistors (7.8 billion equivalent gates), 4 stacks of HBM, 4 DDR5 interfaces supporting 2 DIMMs per channel, and 104 lanes of PCIe Gen5.
Quick Facts
- Abbreviation
- EPI
- Country
- European Union (10 EU countries)
- Ministry
- EU Horizon 2020, EuroHPC JU
- Established
- December 2018
- Disestablished
- March 2026
- Budget
- €150,000,000
Facts from the source article.
Lore & Background
The EPI project was divided into two phases under Specific Grant Agreements: Phase 1 (SGA1) ran from December 2018 to December 2021 under Horizon 2020, and Phase 2 (SGA2) ran from January 2022 to March 2026 under the EuroHPC Joint Undertaking. The consortium was built and selected under the Horizon 2020 Framework Partnership Agreement (FPA: 800928) and consisted of 30 institutions from 10 European countries, led by Bull in France. In 2019, the company SiPearl was incorporated to commercialize the EPI general-purpose processor (GPP) technology.
The project developed GPPs using ARM architecture and accelerators based on RISC-V ISA, targeting integration in Bull Sequana systems. Hardware design was supported by collaborative hardware/software co-design and a full-stack software ecosystem. Application target domains included automotive (Phase 1) and AI (Phase 2). The project concluded in 2026 with significant outcomes, including the demonstration of the Rhea1 chip and the EPAC test-chip family.
Reader's Guide
The European Processor Initiative's significance lies in its strategic goal of establishing European technological sovereignty in high-performance computing. By developing both general-purpose processors (ARM-based) and accelerators (RISC-V-based), EPI created a dual-pathway approach to reduce dependence on non-European technology. The project's most tangible legacy is the Rhea1 processor, featuring 80 Arm Neoverse V1 cores and over 61 billion transistors, which was selected for Jupiter, Europe's first exascale supercomputer. Its successor, Rhea2, was chosen for Alice Recoque, the second European exascale supercomputer. The EPAC accelerator test-chip family, including the Vector Processing Unit, Stencil/Tensor accelerator, and Variable Precision Unit, was silicon-proven and its IPs reused in subsequent EuroHPC projects like DARE. The project also produced hardware-related IPs (Menta Origami EDA tool, ZeroPoint DenseMem memory compression, Kalray KVX accelerators) and released the VPSim virtual prototyping simulator as open source. Several European projects (EUPILOT, EUPEX, DARE) are using EPI results to further European technology sovereignty.
Did You Know?
- SiPearl, the company formed to commercialize EPI's GPP technology, was incorporated in 2019.
- The EPAC 1.5 test chip was silicon-proven and included a Vector Processing Unit, Stencil/Tensor accelerator, and Variable Precision Unit.
- The project organized two editions of the EPI Forum (Barcelona 2024, Paris 2025) and a Codesign Workshop in Heraklion in 2025.
Strategic Mission and Two-Phase Roadmap
The European Processor Initiative was conceived as a continental-scale effort to reduce Europe's reliance on foreign silicon for its most demanding computational workloads. Its stated ambition was to design a new family of low-power microprocessors and accelerators capable of serving extreme-scale computing, high-performance computing, Big Data analytics, edge deployments, and emerging fields such as artificial intelligence and automotive systems. Beyond raw performance, the project carried a deeper geopolitical purpose: to anchor European technological autonomy and sovereignty by placing home-grown processors at the heart of future supercomputers, thereby enabling the continent to tackle societal challenges, accelerate digital transformation, and compete credibly in the global HPC market. The work was structured across two legally distinct phases. Phase 1 ran from December 2018 through December 2021 under the Horizon 2020 programme, while Phase 2, launched in January 2022 and scheduled to close in March 2026, was funded through the EuroHPC Joint Undertaking. Both phases received co-funding from the European Union, participating Member States, national Financing Authorities, and the internal budgets of the partner organisations themselves.
Architecture Choices and Industrial Ecosystem
The consortium made two pivotal architecture decisions that shaped the entire project. General-purpose processor cores were built on the ARM instruction set, while the accelerator line adopted the open RISC-V ISA. Both were intended for integration into Bull Sequana supercomputer systems. To give the technology a commercial path, a dedicated company called SiPearl was incorporated in 2019, serving as the industrial arm tasked with bringing the general-purpose processor technology to market. A defining methodological choice was the emphasis on collaborative hardware-software co-design. Rather than treating silicon and software as separate workstreams, the team developed a comprehensive full-stack software ecosystem alongside the hardware, ensuring that both the general-purpose cores and the accelerators would be supported end-to-end. Validation was grounded in real workloads: flagship scientific applications were used to exercise the designs, and specific target domains were selected for each phase—automotive in Phase 1 and artificial intelligence in Phase 2—so that the new architectures could be stress-tested against application requirements that both benefited from and informed the architectural choices.
Silicon Proofs and Exascale Selections
By the close of Phase 2 in 2026, the project had produced tangible silicon results. SiPearl showed the Rhea1 chip executing a full high-performance computing software stack on a test board. The system booted a standard Linux distribution, interfaced with high-bandwidth memory and its associated performance monitoring counters, and successfully ran both the HPL and STREAM benchmarks—evidence that the complete Rhea1 ecosystem was ready for exascale workloads. The chip packs 80 ARM Neoverse V1 cores, over 61 billion transistors, four stacks of built-in HBM, four DDR5 interfaces, and 104 lanes of PCIe Gen5 connectivity. Jupiter, Europe's first exascale supercomputer, was chosen to implement Rhea1, while its successor, Rhea2, was selected for Alice Recoque, the continent's second exascale machine. In parallel, the EPAC test-chip family gathered a Vector Processing Unit led by SemiDynamics and BSC, a Stencil/Tensor accelerator led by Fraunhofer, and a Variable Precision Unit led by CEA. The EPAC 1.5 test chip reached silicon, and the VEC accelerator was mapped onto a multi-FPGA system built on the MEEP infrastructure.
Consortium Structure and Enduring Legacy
EPI was not a single company's project but a broad European collaboration. The consortium, selected under the Horizon 2020 programme's ICT-42-2017 call, brought together 30 institutions spanning 10 European countries, with France's Bull serving as the coordinating lead. The group included companies, universities, and research organisations, and it operated as a non-legal entity rather than a formal corporate structure. Throughout its life, the consortium organised and participated in numerous community events. The most prominent were two editions of the EPI Forum—held in Barcelona in 2024 and Paris in 2025—and the EPI Codesign Workshop in Heraklion in 2025. The project's influence extends well beyond its 2026 conclusion. Several European follow-on projects, including EUPILOT, EUPEX, and DARE, are building directly on EPI's results to advance the continent's push toward technology sovereignty. Additional hardware-related intellectual property developed during the project—such as the Menta Origami EDA tool, ZeroPoint DenseMem memory compression, and Kalray KVX accelerators—did not reach silicon but remain available for reuse. CEA-List also released VPSim, a virtual prototyping simulator, as open source for early software-hardware co-validation and performance exploration.
Frequently Asked Questions
What is the European Processor Initiative?
EPI is a pan-European program that charts a course for designing next-generation low-power processors and accelerators. It brings together 30 institutions across 10 countries, with France's Bull steering the consortium, to build the silicon backbone for future European supercomputers.
What kind of hardware does EPI target?
The initiative focuses on energy-efficient processor and accelerator families aimed at exascale-class computing, big-data workloads, edge deployments, and emerging fields such as AI and automotive systems.
Who is behind the European Processor Initiative and how is it funded?
The project is led by Bull, a French company, and draws on a consortium of 30 organizations spread over 10 European nations. Phase 1 was funded under Horizon 2020, while Phase 2 runs through the EuroHPC Joint Undertaking.
What are the phases and timeline of EPI?
Phase 1 ran from December 2018 through December 2021, and Phase 2 spans January 2022 to March 2026. Together the two phases cover roughly seven and a half years of roadmap development and design work.
Why is the European Processor Initiative important?
EPI is a strategic push to give Europe its own sovereign path in microprocessor and accelerator design, reducing dependence on non-European suppliers. By powering future European exascale machines, it aims to let the continent tackle major scientific and societal challenges while staying competitive in the global HPC market.
More in Microprocessors 1-24
Spotted an error? Know more?
Reader corrections go straight into our review queue. Suggest an edit · How this site is sourced
