10U Open Compute Server for High Performance Acceleration
The increasing demand for high-performance compute in AI, machine learning, and cloud environments requires scalable and power-efficient server platforms. This case study showcases the development of a 10U Open Compute Project (OCP) Server, engineered to support high-density accelerator workloads with next-generation interconnects and advanced thermal solutions.
The system is designed to deliver optimal performance for compute-intensive applications while maintaining reliability, scalability, and power efficiency across hyperscale deployments.
“High-performance server design is a balance of compute density, thermal efficiency, and scalable architecture.”
Aritrax Engineering Team
Embracing High-Performance Acceleration
The server architecture supports up to six OCP Accelerator Modules (OAM) with OAM v1.5 accelerators (~900W each), enabling extreme compute density within a compact 10U form factor. The platform is built on an AMD EPYC processor-based motherboard, providing robust processing capability and high memory bandwidth.
With support for up to eight DDR5 ECC DIMMs per node, the system ensures data integrity and performance for mission-critical workloads. Expansion and connectivity are enabled through two PCIe Gen 5 x16 slots and one OCP NIC 3.0 slot, allowing seamless integration of high-speed accelerators and networking components.
To meet modern storage requirements, the server includes two M.2 slots (PCIe Gen5 x4) and two 2.5-inch hot-swappable SATA600 drive bays, offering flexibility and serviceability in data center environments.
Engineering Responsibilities & Challenges
The project involved end-to-end ownership across multiple engineering domains, including project management, architecture design, hardware design, CAD layout, and system bring-up. Interface testing was performed using a 3rd party ASRock Genova AMD motherboard to ensure compatibility and reliability.
Key challenges addressed during development included achieving stable PCIe Gen 5 performance and implementing an efficient thermal cooling solution. These challenges were overcome through an integrated negative-pressure direct-to-chip liquid cooling system and an Open Rack V2 power supply, ensuring sustained performance under high load conditions.
Applications
This OCP server platform is optimized for a wide range of enterprise and data-center use cases:
1. AI & Machine Learning: High-density accelerator support for deep learning training
2. High-Performance Computing (HPC): Optimized compute for scientific and simulation workloads
3. Cloud & Hyperscale: Scalable deployments with high compute density and power efficiency
