Chipmaker AMD said it is expanding its partnership with technology company Hewlett Packard Enterprise (HPE) to build a new (artificial intelligence) AI server platform.
HPE will be among the first to use the chipmaker’s “Helios” rack-scale AI system, which includes a special HPE networking switch developed with Broadcom and software for fast, high-capacity data connections.
“‘Helios’ combines AMD processors, GPUs, networking chips, and the AMD ROCm software to create a complete system designed for performance, efficiency, and flexibility,” said Dr. Lisa Su, chair and CEO of AMD. “We are bringing together AMD’s technologies and HPE’s system design to deliver an open AI platform that improves efficiency and scalability for our customers,”
The Helios system can deliver up to 2.9 exaFLOPS of FP4 performance per rack, using the chipmaker’s Instinct MI455X GPUs, next-generation EPYC “Venice” processors, and AMD Pensando Vulcano network cards. All parts work together through the ROCm software, which supports AI and high-performance computing tasks.
Built on the Open Compute Project (OCP) Open Rack Wide design, Helios is meant to make deploying large AI systems faster and more flexible.
“With AMD ‘Helios’ and our HPE scale-up networking solution, we are giving cloud providers faster setups, more flexibility, and lower risks when scaling AI computing,” said Antonio Neri, president and CEO at HPE.
The system’s switch and software, developed with Broadcom, are optimized for AI workloads using the Ultra Accelerator Link over Ethernet (UALoE) standard. HPE plans to make Helios available worldwide in 2026.