Eviden, the product brand of the Atos group specializing in advanced computing, has introduced the BullSequana XH3500, a new generation of supercomputer designed for the convergence of HPC (High-Performance Computing) and artificial intelligence. The company positions it as the technological foundation for future supercomputing centers and the so-called AI factories, facilities dedicated to training and deploying large-scale AI models continuously.
The announcement comes after more than two decades of Eviden/Atos leadership in supercomputing with the BullSequana family. The new XH3500 is presented as the natural evolution of the BullSequana XH3000, but with a significant leap in power density, energy efficiency, and architectural flexibility. These elements are crucial in a context where data centers seek higher performance without increasing power consumption or rack space.
A platform designed for HPC + AI convergence
Eviden’s proposal is based on a reality: the line between traditional scientific simulation and AI workloads has blurred. The same centers that run fluid dynamics codes, climate models, or material simulations now train language models, vision systems, and large neural networks.
The BullSequana XH3500 is introduced precisely as a convergent platform capable of managing both worlds. It’s not just about “adding more GPUs,” but integrating multiple components within a single chassis:
- nodes optimized for numerical simulation and physical models,
- nodes for massive AI training,
- and low-latency interconnection networks that enable both workloads to feed into each other.
Eviden frames the XH3500 within a broader ecosystem of software, services, and expert support, covering everything from design and installation to application optimization and daily operations. The goal is to reduce the time between hardware investment and scientific or business results.
More power in the same space: up to 80% more energy per square meter
One of the most striking data points of the new system is its power density and cooling capacity. Compared to its predecessor, the BullSequana XH3000, the XH3500 can:
- deliver over 80% additional electrical power per square meter,
- offer 30% more cooling capacity per kilowatt,
all while maintaining the same rack footprint.
Practically, this means a data center can house many more CPUs and GPUs in the same physical space without expanding the room or radically modifying cooling infrastructure. For large supercomputing centers, often limited by space and power capacity, this increase in density is a decisive factor.
The company emphasizes that the platform is designed to scale beyond exascale, meaning systems capable of exceeding one exaflop of floating-point performance. The XH3500 will, in fact, serve as the foundation for future European exascale supercomputers, such as the French project Alice Recoque, with a strong focus on sovereign AI and high-impact scientific simulation.
Open modular architecture: CPU, GPU, and accelerators à la carte
Another pillar of the BullSequana XH3500 is its modular and open design. Unlike closed solutions, the system allows each customer to mix and match different processor, acceleration, and networking technologies based on their needs:
- various high-performance CPU families,
- multiple generations of GPU and AI accelerators,
- and different options for high-speed interconnection.
This approach aims at two clear goals:
- Avoid vendor lock-in and allow for adopting the best available technology at any time, whether CPUs, GPUs, or NICs.
- Protect long-term investment by enabling the integration of new processors or networks as they emerge, without replacing the entire infrastructure.
For supercomputing center managers and major cloud operators, this flexibility is particularly relevant in a chip market characterized by rapid cycles and new architectures, from specialized GPUs to future quantum accelerators.
Direct liquid cooling and “fan-less” operations
The increased rack power density inevitably leads to more heat to evacuate. Eviden addresses this challenge with the fifth generation of its Direct Liquid Cooling (DLC) technology.
The BullSequana XH3500 is a 100% “fan-less” rack system, based on:
- Fifth-generation DLC,
- use of lukewarm water with inlet temperatures around 40°C,
- and a design that allows all critical components to be cooled via water circuits, reducing reliance on traditional air conditioning.
This setup opens the door to residual heat reuse strategies (e.g., for district heating or industrial processes) and free-cooling configurations, where climate conditions minimize mechanical cooling needs. The result is a significant reduction in Total Cost of Ownership (TCO) and environmental impact, key aspects for centers needing to justify their energy footprint to funders and regulators.
Additionally, the platform is supported by software tools for intelligent energy and resource management, capable of real-time monitoring, load balancing, and performance optimization based on efficiency goals or user priorities.
Ready for “AI factories” and the next wave of technology
According to the company, the BullSequana XH3500 is designed to meet the needs of current and future “AI factories”: infrastructures where AI models are trained, deployed, and served continuously, often combining data from physical simulations and digital twins.
The hybrid HPC-AI architecture of the XH3500 aims to:
- efficiently run large scientific simulations,
- train state-of-the-art AI models on that data,
- and support workflows where the model learns, validates with simulations, and iteratively self-tunes.
Eviden also sees the system as a platform ready to integrate emerging technologies, including quantum computing, as these mature and become available as specialized accelerators within the BullSequana ecosystem.
Market voice: flexibility and real-world use cases
From the user perspective, one of the most valued aspects is the combination of AI capacity with a strong focus on traditional HPC. Sector analysts highlight that many scientific and technical centers are seeking exactly this convergence: continuing to run high-performance simulation codes while enriching them with AI models that help analyze results, reduce computation times, or explore broader design spaces.
The modularity of the BullSequana XH3500 allows tailoring configurations to each case: from climate and weather clusters to platforms focused on drug discovery, seismic analysis, energy, or advanced industrial design, always with the possibility to incorporate new generations of CPUs, GPUs, and networks without completely rebuilding the data center.
Frequently Asked Questions about Eviden’s BullSequana XH3500
What is the BullSequana XH3500, and what workloads is it designed for?
The BullSequana XH3500 is Eviden’s new supercomputing platform, designed for the convergence of HPC and artificial intelligence. It targets supercomputing centers, “AI factories,” and large organizations that need to combine scientific simulation, data analysis, and large-scale AI model training in a single system, offering high power density and a strong focus on energy efficiency.
How does the BullSequana XH3500 differ from the previous BullSequana XH3000 model?
Compared to the XH3000, the XH3500 provides over 80% more electrical power per square meter and about 30% more cooling capacity per kilowatt, while maintaining the same physical rack footprint. This results in higher compute density within the same space, better utilization of existing infrastructure, and a more adaptable platform to support configurations with numerous GPUs and AI accelerators.
How does the modular architecture of the BullSequana XH3500 help avoid vendor lock-in?
Its open and modular design allows combining different types of CPUs, GPUs, AI accelerators, and network interconnects. Each customer can select the best-fit technologies and upgrade components gradually as new processor or network generations become available. This flexibility reduces dependence on a single vendor and safeguards long-term investment in the data center.
What sustainability and cost benefits does the BullSequana XH3500’s direct liquid cooling offer?
The fifth-generation direct liquid cooling cools all critical components with lukewarm water, eliminating rack fans and reducing the need for conventional air conditioning. This significantly lowers cooling energy consumption, facilitates residual heat reuse strategies, and helps meet sustainability goals. Reduced energy costs and higher power density also lower the Total Cost of Ownership (TCO) over the system’s lifetime.
via: eviden