Barcelona Supercomputing Center
The Barcelona Supercomputing Center – Centro Nacional de Supercomputación (BSC-CNS) is Spain's national center for supercomputing, established in 2005 as a multidisciplinary research institution focused on high-performance computing (HPC), located at the Barcelona Polytechnic University campus in Barcelona, Catalonia.[1][2] It serves as a hub for scientific and industrial innovation, providing advanced computational resources to researchers across Europe and beyond, while promoting the development and adoption of HPC technologies in fields such as computer sciences, life sciences, earth sciences, engineering applications, and computational social sciences and humanities.[1] The center manages the MareNostrum series of supercomputers, with MareNostrum 5—a pre-exascale system operational since late 2023—representing a major European investment in scientific infrastructure at €202 million, featuring a peak performance of 314 petaflops across its general-purpose and accelerated partitions equipped with Intel Sapphire Rapids processors and NVIDIA Hopper GPUs.[3][4][5] BSC-CNS traces its origins to the Centre Europèu de Paral·lelisme de Barcelona (CEPBA), founded in 1991 by the Universitat Politècnica de Catalunya (UPC) with support from Spanish and Catalan research agencies, which evolved through collaborations like the Centre de Supercomputació de Catalunya (CESCA) from 1995 to 2000 and a 2000 partnership with IBM.[2] Officially created in 2005 through a collaboration between the Spanish Ministry of Education and Science, the Generalitat de Catalunya, and UPC, the center launched with MareNostrum 1, Europe's then-most powerful supercomputer at 42.35 teraflops, marking a milestone in European HPC infrastructure.[2] Subsequent upgrades, including MareNostrum 3 in 2012 (1.1 petaflops) and MareNostrum 4 in 2017 (13.9 petaflops), solidified its leadership, with MareNostrum 5 achieving dual entries in the TOP500 list's top 20 in 2023 as the only European supercomputer to do so.[6][7] Under the direction of Mateo Valero since its inception, BSC-CNS serves as a hosting site for supercomputers of the EuroHPC Joint Undertaking, manages the Spanish National High-Performance Computing Network (RES), and is a founding member of the Partnership for Advanced Computing in Europe (PRACE), facilitating Tier-0 access to petabyte-scale storage and high-speed networks for numerous projects annually in areas like climate modeling, biomedicine, and astrophysics.[1][6] In 2025, the center continues to advance emerging technologies, including the deployment of Spain's first fully European quantum computer and the establishment of an AI Factory as one of seven such facilities in Europe, emphasizing energy-efficient architectures like RISC-V and supporting national priorities in AI and multilingual processing.[8][9][10]History
Founding and Early Developments
The origins of the Barcelona Supercomputing Center (BSC) trace back to the Centre Europèu de Paral·lelisme de Barcelona (CEPBA), established in 1991 at the Universitat Politècnica de Catalunya (UPC). CEPBA was formed by aggregating expertise from various UPC departments, including the Computer Architecture Department (DAC), to advance efficient computing technologies for academic and industrial applications. It was initially sponsored by the Spanish National Plan for Research and Development (CICYT) and the Catalan Interdepartmental Commission for Research and Technological Innovation (CIRIT), focusing on parallelism and high-performance computing innovations.[2] This foundation evolved through subsequent collaborations, such as the CEPBA-IBM Research Institute launched in 2000, a four-year partnership emphasizing deep computing and architecture research, and earlier efforts like the C4 consortium with the Catalan Supercomputing Center (CESCA) from 1995 to 2000. These initiatives, including successful public-private partnerships like those involving CIRIT, laid the groundwork for a national facility. In 2005, the BSC was officially created as Spain's national supercomputing center through a joint agreement between the Spanish Ministry of Education and Science, the Generalitat de Catalunya, and UPC, marking a transition from regional efforts to a centralized infrastructure supporting broader scientific needs.[2][11] A pivotal early development was the launch of the first MareNostrum supercomputer in 2005, housed in the Torre Girona chapel at UPC. Built by IBM, it featured 4,812 PowerPC 970 processors across 2,406 blades, delivering a peak performance of 42.3 teraflops, making it Europe's most powerful supercomputer at the time. The BSC's initial emphasis was on providing high-performance computing (HPC) resources to address national and European research demands in science and engineering, fostering accessibility for researchers through hosted services and knowledge transfer. Early partnerships with the Spanish government, Catalan authorities, and UPC ensured funding, operational support, and physical hosting, establishing BSC as a cornerstone for advanced computational capabilities in Spain.[2][12]Key Milestones and Growth
Following its establishment in 2005, the Barcelona Supercomputing Center (BSC) experienced significant institutional expansion, marked by key technological upgrades and recognitions that solidified its position as a leading European research hub. In 2006, BSC introduced MareNostrum 2, enhancing computational capacity to 94.2 teraflops, which positioned it as Europe's top supercomputer at the time.[6] This upgrade supported broader scientific applications and set the stage for subsequent growth. By 2010, BSC became a founding and hosting member of the Partnership for Advanced Computing in Europe (PRACE), enabling trans-national access to high-performance computing resources across the continent.[13] The center's research excellence was formally acknowledged in 2011 with the Severo Ochoa Centre of Excellence accreditation from the Spanish Ministry of Science and Innovation, recognizing BSC's international impact in computational sciences.[14] This distinction was renewed in 2015 for the 2016-2020 period and again in 2022 for 2023-2027, each accompanied by multi-year funding to support advanced research initiatives.[15] Concurrently, infrastructure advancements continued with the deployment of MareNostrum 3 in 2012, achieving 1.1 petaflops of peak performance and expanding BSC's role in large-scale simulations for fields like climate modeling and biomedicine.[6] In 2017, MareNostrum 4 launched with 13.9 petaflops, further elevating BSC's ranking among global supercomputing facilities and integrating emerging technologies such as GPU acceleration.[6] BSC's growth accelerated through its involvement in the EuroHPC Joint Undertaking, established in 2018 to procure and deploy world-class supercomputers across Europe, with BSC selected as a host site for pre-exascale systems.[16] This culminated in the 2023 launch of MareNostrum 5, Europe's largest scientific infrastructure investment in Spain at 202 million euros, delivering 314 petaflops peak performance.[4] The system features a general-purpose partition at 45.9 petaflops using Intel Sapphire Rapids processors and an accelerated partition at 260 petaflops powered by NVIDIA H100 GPUs, alongside 248 petabytes of storage for data-intensive workloads.[3] Organizational expansion paralleled these developments, with staff growing from approximately 60 in 2005 to nearly 1,400 by 2025, reflecting increased demand for interdisciplinary expertise.[17] Annual budgets expanded from an initial 5.5 million euros to over 34 million euros by 2018, bolstered by competitive grants and partnerships.[18] By 2024, BSC secured an additional 100 million euros from Next Generation EU funds to advance AI, quantum, and sustainability research.[19] In 2024-2025, BSC integrated quantum computing capabilities, testing Spain's first fully European quantum system—a digital gate-based computer developed by Qilimanjaro Quantum Tech and GMV—directly on MareNostrum 5, marking a pivotal step toward hybrid classical-quantum infrastructures.[20] This integration, part of the EuroHPC Quantum Computers Spain initiative, enhances BSC's leadership in next-generation computing paradigms.[21] In November 2025, BSC celebrated its 20th anniversary, highlighting its evolution into a key European hub for supercomputing, AI, and technological sovereignty.[17]Facilities and Infrastructure
Supercomputing Systems
The Barcelona Supercomputing Center's primary high-performance computing infrastructure centers on the MareNostrum series, which has evolved through successive generations to deliver escalating computational power while prioritizing energy efficiency. Launched in 2005, MareNostrum 1 featured 4,812 IBM PowerPC 970FX processors running at 2.2 GHz across 2,406 JS20 blades, achieving a peak performance of 42.3 teraflops (TFLOPS) with 9.6 terabytes (TB) of main memory and 236 TB of disk storage; it was notable for its innovative water-cooling system housed in a transparent enclosure.[12] Subsequent upgrades marked significant advancements: MareNostrum 2, deployed in 2008, utilized 10,240 IBM PowerPC 970MP processors at 2.3 GHz in 2,560 JS21 blades, boosting peak performance to 94.21 TFLOPS with 20 TB of main memory and over 480 TB of combined disk storage.[22] By 2012, MareNostrum 3 shifted to Intel Sandy Bridge architecture with 48,896 cores across 3,056 nodes, reaching 1.1 petaflops (PFlops) peak while incorporating Infiniband FDR10 networking for enhanced interconnectivity.[23] MareNostrum 4, operational from 2017, employed Intel Xeon Platinum processors in 3,456 nodes for the general-purpose partition, delivering 11.15 PFlops peak (expandable to 13.9 PFlops with hybrid additions), 384.75 TB of memory, and improved power efficiency through Lenovo SD530 racks and Intel Omni-Path interconnects.[24] These iterations progressively enhanced scalability, reduced energy per flop, and integrated hybrid computing elements to support diverse workloads. MareNostrum 5, inaugurated in 2023 as a pre-exascale system under the EuroHPC Joint Undertaking, represents the pinnacle of this evolution with a total peak performance of 314 PFlops, comprising a general-purpose partition (GPP) and an accelerated computing partition (ACC). The GPP consists of 6,480 nodes (including 72 high-bandwidth memory variants), each equipped with dual Intel Xeon Sapphire Rapids 8480+ processors (112 cores per node at 2 GHz), 256 GB DDR5 memory (up to 1 TB on select nodes), and 960 GB NVMe storage, yielding 45.9 PFlops peak across 725,760 cores and 1,775 TB total memory.[3] The ACC features 1,120 nodes, each with dual Intel Xeon Sapphire Rapids 8460Y+ processors (80 cores per node at 2.3 GHz) and 4 NVIDIA H100 Hopper GPUs (64 GB HBM2e each), delivering 260 PFlops peak for AI and simulation-intensive tasks, with the overall system supported by Bull Sequana XH3000 and Lenovo ThinkSystem architectures running Red Hat Enterprise Linux.[25][16] Storage infrastructure includes a 248 petabyte (PB) parallel file system using SSD/Flash and hard disks, offering 1.2 terabytes per second (TB/s) write and 1.6 TB/s read bandwidth, alongside tape-based archival up to 400 PB.[3] Supporting infrastructure enables seamless high-speed data movement and emerging hybrid paradigms, including an InfiniBand NDR200 network connecting over 7,800 nodes at up to 200 Gb/s for low-latency communication (with variations by partition: 100 Gb/s for GPP nodes, 800 Gb/s for ACC nodes).[26] In 2025, integration of a quantum accelerator—Spain's first such system, featuring initial superconducting Transmon qubit chips with 10 to 20 qubits in a cryogenic setup developed with 100% European technology—expands capabilities for hybrid classical-quantum computing, connected via dedicated interfaces to the main cluster.[20][27] This quantum module, part of the Quantum Spain initiative, supports exploratory workloads in optimization and simulation. MareNostrum 5 also includes Next Generation Technology partitions, such as the NGT GPP with NVIDIA Grace Arm-based CPUs, to test future architectures.[3] MareNostrum 5 includes dedicated partitions for urgent computing, enabling rapid allocation of resources—such as over 200 GPUs in recent tests—for time-critical simulations during emergencies like earthquakes or pandemics, as demonstrated in Mexico's 2025 national earthquake drill where protocols were validated for immediate response.[28] Energy efficiency defines MareNostrum 5 as Europe's greenest supercomputer, powered entirely by renewable sources with a power usage effectiveness (PUE) below 1.08, ensuring high efficiency in energy use.[29][30] Cooling combines air-based systems for auxiliary components with advanced liquid cooling, including direct-to-node warm water and rear-door heat exchangers, to manage up to 20 megawatts (MW) while reusing waste heat for district heating—aligning with the European Green Deal's sustainability objectives and minimizing environmental impact.[30]| System | Deployment Year | Peak Performance | Key Processors | Notable Efficiency Features |
|---|---|---|---|---|
| MareNostrum 1 | 2005 | 42.3 TFLOPS | IBM PowerPC 970FX | Water-cooling enclosure |
| MareNostrum 2 | 2008 | 94.21 TFLOPS | IBM PowerPC 970MP | Enhanced blade density |
| MareNostrum 3 | 2012 | 1.1 PFlops | Intel Sandy Bridge | Infiniband FDR10 networking |
| MareNostrum 4 | 2017 | 11.15 PFlops (13.9 total) | Intel Xeon Platinum | Omni-Path interconnects, hybrid expansion |
| MareNostrum 5 | 2023 | 314 PFlops | Intel Sapphire Rapids / NVIDIA Hopper | PUE below 1.08, renewable power, heat reuse |