Blog written by Celestica, published Apr. 15, 2026

The recent deployment of the “Helma” supercomputer at the Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) marks a pivotal moment for German academic research, but it also signals a bold new era of industry leadership for Celestica. While the system’s massive GPU array often captures the headlines, the project’s success is truly anchored by Celestica’s deep expertise and architectural leadership in AI for the HPC.

MEGWARE, one of Europe’s leading supercomputing specialists, leveraged their expertise in HPC system engineering to integrate and implement the storage infrastructure. They designed and deployed a sophisticated storage ecosystem utilizing Xinnor’s xiRAID and Celestica’s advanced storage platform to power an all-NVMe Lustre parallel file system. This collaboration resulted in a #3 global ranking on the IO500 list.

SC6100 Storage Layer: Eliminating I/O Bottlenecks
At the heart of the Helma project, Celestica’s storage controller serves as the glue that binds this high-performance ecosystem together. By anchoring the architecture with the SC6100 – a high-availability, dual-node all-flash controller – Celestica seamlessly integrates PCIe Gen5 technology and AMD EPYC processors to eliminate the I/O bottlenecks that typically starve power-hungry GPUs. This Storage Bridge Bay (SBB) design acts as the ultimate stabilizer, unifying disparate components into a reliable, redundant system that eliminates single points of failure for long-running AI training sessions. Ultimately, Celestica’s ability to fuse these advanced technologies within a compact half-rack footprint resulted in power and cooling performance well ahead of traditional multi-rack appliances.

Reflecting on the engineering heritage behind these solutions, George Tehrani, field CTO & head of enterprise product strategy, Celestica, noted: “Our role in the Helma project is a testament to our deep product management and engineering expertise in design-led manufacturing. On this project, we were a strategic design partner focused on architecting the core infrastructure that is fueling the AI revolution. Our history in complex data center infrastructure allows us to deliver high-density, mission-critical systems that meet the rigorous demands of the world’s most advanced supercomputing environments.”

Delivering an Open Systems Blueprint for Cost-Effective AI Storage
Beyond raw performance, the collaboration highlights the “commodity economics” and commitment to open systems that Celestica champions. By utilizing high-quality, open components rather than proprietary, locked-in hardware, Celestica empowers customers to embrace disaggregation, decoupling hardware from software to scale their infrastructure more flexibly and cost-effectively. This approach not only benefited the German research community by providing an agile, vendor-neutral environment but also serves as a repeatable blueprint for enterprise customers looking to deploy AI-ready storage more effectively. 

Designed for Mission-Critical Supercomputing
Speaking to these broader market benefits, David Autrey, Head of Enterprise Storage at Celestica, stated: “The success of the Helma project demonstrates a significant advantage for our broader customer base: the ability to achieve world-class, high-availability storage performance using open, efficient architectures. By maximizing GPU utilization and minimizing the physical footprint, we are helping our customers solve the multiple challenges of performance, cost, and sustainability. This case study further demonstrates how Celestica’s platform solutions deliver the agility and efficiency needed to lead in the rapidly evolving AI, Hybrid Cloud and HPC ecosystems.”

Celestica’s technology leadership on the Helma supercomputer deployment serves as a powerful validation of their SC6100 storage platform, delivering record-breaking performance and high-availability reliability within an ultra-dense footprint. By combining the SC6100’s cutting-edge PCIe Gen5 hardware with the high-performance xiRAID solution from key storage software ecosystem partner Xinnor, Celestica has provided a scalable, cost-effective blueprint for the future of AI-driven research. For a deeper look at the technical specifications and the project’s impact on German supercomputing, read the recent blog and explore the detailed Case Study.