From Complexity to Efficiency: Open Hardware Strategy Through Collaboration

Author: Charlotte Lee, ASRock Rack; Stefan Dutila, VP Hardware Engineering, Scaleway Collection: 2025 OCP Global Summit - Expo Hall Sessions

Summary

ASRock Rack and Scaleway co-present their collaboration on a 4U 12-node ORV3-compatible server system running OpenBMC. ASRock Rack is a server hardware company building comprehensive product lines across cloud, enterprise, and edge computing, including high-density storage (U.2, E1.S, E3.S), AI GPU servers (from compact nodes for H environments to liquid-cooled rack-level solutions), CPU multi-node servers, and a broad range of server boards from mini-ITX to multi-socket supporting x86 and ARM architectures.

Scaleway, part of Iliad (the fifth-largest telco in Europe), is a European cloud service provider operating thousands of GPU and general-purpose servers across Europe, expanding to the US. Four years ago, Scaleway began investigating open firmware (OpenBMC) and OCP hardware standards to address operational challenges: SKU proliferation, firmware management complexity, power efficiency, serviceability, and maintenance across a large, diverse installed base. The company evaluated MHS form factors, ORV3 benefits (higher rack-level density, front-access serviceability, power efficiency), and 21-inch chassis (seeing little value in MHS within 19-inch chassis).

The resulting 4U 12-node system features AM5-based nodes, front-serviceable and hot-pluggable design, no central management controller (nodes communicate via I2C buses for cost efficiency), two fan zones serving six nodes each with hot-swappable fans and single-fan-fail redundancy. Scaleway invested dedicated engineering teams to express specifications, challenge ASRock Rack's proposals, and provide hardware/firmware feedback. The system runs OpenBMC and is about to be released as a production product. Scaleway has also accepted co-leading a new OCP work stream called "Scaling AI Clusters at Neoclouds."