BullSequana eXascale Interconnect (BXI v3)
With growing HPC and AI workloads and the advent of Exascale, interconnect networks face mounting pressure. BXI v3 offers a game-changing solution by offloading tasks from GPUs and efficiently managing traffic, boosting utilization and system performance.
As Europe’s leading interconnect player and a founding member of the UltraEthernet Consortium (UEC), BXI v3 delivers an open-standard, sovereign solution tailored for AI and HPC workloads.
BullSequana BXI v3 Benefits
Insights
Earl Joseph
Earl Joseph
The launch of Eviden’s newest BXI marks a significant milestone in the HPC market, one that speaks to the growing demand for innovative network technologies designed to unlock the full potential of HPC and Artificial Intelligence. As the appetite for AI-ready scalable infrastructures continues to grow, Eviden’s unique approach offers a strong perspective, combining seamless plug-and-play capabilities with a deep understanding of market needs. By embracing the most widely used protocol in the market, while extending it with the unique features of application protocol offload to the NIC, Eviden brings to the market a real SmartNIC Network with Plug&Play capabilities for both OpenMPI (HPC) and NCCL/RCCL (AI) highly demanding workloads. This solution is designed to become an essential tool for organizations aiming to enhance the potential of their CPU and GPU power while improving TCO and ease of integration."
Related resources
November 14, 2024
Eviden leads the charge in high-performance computing with new AI-ready scale-out networking offering
BXI v3, the third generation of the only European scale-out networking technology, specifically designed for Artificial Intelligence (AI) and high-performance computing (HPC) workloads, set to be available in the second half 2025.
BullSequana eXascale Interconnect v3
Intelligent Network Management Accelerates GPU Performance in AI-HPC
FAQ
HPC compute power has surged, but networking lags. Despite bandwidth growth, it remains insufficient. Reducing latency and scaling intelligently with open-standard protocols are crucial for future workloads.
Existing networking solutions often create bottlenecks by over-relying on CPUs for data processing, increasing latency and slowing response times. This leads to underutilized GPUs, idle up to 70% of the time, and limits overall throughput. Despite increasing bandwidth, scalability challenges, high CPU overhead, and limited optimization features persist, complicating infrastructure management.”
BXI v3, a fully European technology, natively supports Ethernet and TCP/IP, enabling scalable cluster aggregation. Its Smart NIC offloads GPU communication tasks, reducing idle time and improving ROI through better GPU utilization, lower TCO, and enhanced performance. With integrated RDMA and Atomics support, plus advanced congestion management, it ensures faster, more efficient application execution with plug-and-play simplicity
The general availability of BXI v3 will be 3Q2025. Please reach out to the following experts for any questions you may have.
Contact Us
Get in touch!
Connect with us to discover BXI V3
Get in Touch!