RED-SEA · Project

Faster Network Connections for Supercomputers That Run AI and Big Data

digitalTestedTRL 5

Imagine a highway system connecting hundreds of thousands of computers so they can work together as one giant brain. The current roads are getting too narrow for the growing traffic from AI and big data. RED-SEA upgraded Europe's main supercomputer highway — called BXI — to handle exascale traffic, meaning a billion billion calculations per second. They also built a bridge so these supercomputers can talk smoothly to regular office networks running on Ethernet.

By the numbers

hundreds of thousands

nodes the interconnect architecture scales to

consortium partners

countries represented

Top 500

supercomputer systems already using BXI

project deliverables produced

42%

industry ratio in consortium

The business problem

What needed solving

Companies running supercomputers for AI training, scientific simulation, and big data analytics are hitting network bottlenecks. As workloads grow toward exascale, the interconnect — the network fabric linking thousands of compute nodes — becomes the limiting factor. Existing networks struggle with congestion, lack virtualization support, and cannot bridge efficiently to standard Ethernet infrastructure.

The solution

What was built

The project delivered 10 outputs including a working BXI to Ethernet bridging demonstrator with validation report. They designed and tested next-generation interconnect architecture features through FPGA testbeds, mathematical modeling, and simulation — covering adaptive routing, congestion management, virtualization, and security enhancements for the BXI interconnect.

Audience

Who needs this

HPC data center operators scaling toward exascaleCloud providers offering high-performance computing servicesAutomotive and aerospace companies running large-scale simulationsAI companies training distributed large language modelsNational supercomputing centers upgrading their infrastructure

Business applications

Who can put this to work

High-Performance Computing Services

enterprise

Target: HPC data center operators and cloud providers offering compute-as-a-service

If you are an HPC data center operator dealing with network bottlenecks as you scale AI and analytics workloads — this project developed next-generation BXI interconnect features tested on FPGA testbeds that scale to hundreds of thousands of nodes. The BXI to Ethernet bridging demonstrator means your supercomputer clusters can connect seamlessly to standard enterprise networks. A consortium of 12 partners across 6 countries validated these improvements.

Automotive & Aerospace Engineering

enterprise

Target: Engineering firms running large-scale simulations and digital twins

If you are an engineering company running crash simulations, aerodynamics modeling, or digital twins that push your compute infrastructure to its limits — this project improved congestion handling, adaptive routing, and collective operations in supercomputer networks. These upgrades mean your simulation jobs finish faster with fewer communication delays between compute nodes. The technology builds on BXI, already featured in Top 500 supercomputer systems.

AI and Machine Learning

mid-size

Target: Companies training large AI models requiring distributed computing

If you are an AI company struggling with the network overhead of training large models across many GPUs — RED-SEA designed interconnect improvements specifically for converging HPC, big data analytics, and AI workloads. The project developed enhanced network resource management including virtualization support, letting you run multiple AI training jobs simultaneously without them interfering with each other. The 42% industry ratio in the consortium ensured real-world requirements drove the design.

Frequently asked

Quick answers

What would it cost to adopt this interconnect technology?

RED-SEA built on BXI, a commercial interconnect product from Bull (now Eviden/Atos). Pricing for BXI-based network hardware is available through Eviden's sales channels. The project advanced the next-generation architecture, so commercial availability of the specific RED-SEA improvements depends on Eviden's product roadmap.

Can this scale to our data center size?

The architecture was designed to scale to hundreds of thousands of nodes, targeting exascale systems. The features were tested through mathematical modeling, simulation, and FPGA testbeds. BXI is already deployed in Top 500 supercomputer systems, confirming production-grade scalability for the base technology.

What is the IP situation and how can we license this?

Bull SAS (now Eviden, part of Atos Group) coordinated the project and owns the core BXI technology commercially. As a publicly funded RIA project, results follow EU open access rules, but the commercial interconnect product is proprietary. Contact Eviden directly for licensing and procurement of BXI-based products.

Does this work with our existing Ethernet infrastructure?

Yes — one of the key deliverables was a BXI to Ethernet bridging demonstrator, specifically designed for seamless communication between HPC clusters and standard Ethernet networks. This means you can integrate supercomputing resources into your existing network without replacing everything.

How does this compare to competing technologies like InfiniBand?

BXI is the key European alternative to InfiniBand for HPC interconnects. RED-SEA focused on making BXI competitive at exascale by adding features like adaptive routing, congestion management, and virtualization support. Based on available project data, specific benchmark comparisons were not published in the summary materials.

When will the next-generation features be commercially available?

The project ran from 2021 to 2024 and completed its research and testing phase. Eviden controls the commercial roadmap for BXI products. Based on available project data, the timeline for incorporating RED-SEA innovations into production BXI hardware depends on Eviden's product cycles.

Consortium

Who built it

The RED-SEA consortium of 12 partners from 6 countries (Switzerland, Germany, Greece, Spain, France, Italy) has a strong industrial backbone with 42% industry ratio — 5 industrial partners including 4 SMEs. The project is led by Bull SAS (now Eviden/Atos), which is the manufacturer of BXI and one of Europe's largest technology companies. This means the path from research to commercial product is unusually direct: the company that makes and sells the interconnect hardware is the one leading the research. The mix of 3 universities and 4 research organizations provided the scientific muscle for architectural design and mathematical modeling, while the industrial partners ensured the results target real market needs.

BULL SASCoordinator · FR
FORSCHUNGSZENTRUM JULICH GMBHparticipant · DE
COMMISSARIAT A L ENERGIE ATOMIQUE ET AUX ENERGIES ALTERNATIVESparticipant · FR
EXACT LAB SRLparticipant · IT
IDRYMA TECHNOLOGIAS KAI EREVNASparticipant · EL
EIDGENOESSISCHE TECHNISCHE HOCHSCHULE ZUERICHparticipant · CH
EXTOLL GMBHparticipant · DE
UNIVERSIDAD DE CASTILLA - LA MANCHAparticipant · ES
EXASCALE PERFORMANCE SYSTEMS - EXAPSYS IKEparticipant · EL
PARTEC AGthirdparty · DE
UNIVERSITAT POLITECNICA DE VALENCIAparticipant · ES
ISTITUTO NAZIONALE DI FISICA NUCLEAREparticipant · IT

How to reach the team

Bull SAS (Eviden/Atos Group), France — contact through Eviden's HPC division for commercial inquiries about BXI interconnect technology

Order a report on this consortium See BULL SAS profile →

Next steps

Talk to the team behind this work.

Want to know if RED-SEA's interconnect improvements fit your data center upgrade plans? SciTransfer can arrange an introduction to the right technical contacts in the consortium.

Order a report on this topic More in Digital & ICT