At HPE’s Discover Barcelona 2025 event today, the company said it is offering the AMD “Helios”AI rack-scale architecture with Ethernet, intended to speed up deployment of at-scale artificial intelligence training and inferencing for cloud service providers.
The solution utilizes HPE Juniper Networking hardware and software along with Broadcom’s Tomahawk 6 networking chip and is based on the open Ultra Accelerator Link over Ethernet (UALoE)standard.
HPE said the solution is capable of supporting traffic from trillion parameter model training, high inference throughput and massive model sizes.
HPE said it will be one of the first companies to offer Helios, which is based on specifications from the Open Compute Project and are optimized for power delivery, advanced liquid cooling and high serviceability as needed for AI systems, according to the company.
Features include:
- Rack-scale performance: Connecting 72 AMD Instinct MI455X GPUs per rack,the solution delivers 260TB/second of aggregated scale-up bandwidth and up to 2.9 AI exaflops of FP4 performance. The system provides 31TB of HBM4 and 1.4PB/second of memory bandwidth.
- Standards-based Ethernetscale-up switch and software: HPE is extending its “networks for AI” solutions with a scale-up Ethernet switch and software built for Helios, developed in collaboration with Broadcom. The switch leverages HPE’s AI-native automation capabilities to simplify network operations for faster deployment and cost savings.
- Open standards and innovation: Built on OCP’s Open Rack Wide (ORW) specifications, the double-wide design of the Helios architecture offers liquid cooling designed for ongoing maintenance, HPE said. Helios also leverages AMD ROCm software and AMD Pensando networking technology. HPE said the HPE Juniper Networking switch minimizes vendor lock-in and enables swifter feature updates.
- HPE will offer the Helios solution worldwide in 2026.
Broadcom is working with HPE to develop the HPE Juniper Networking scale-up switch for Helios. “Broadcom is proud to take part in this collaboration to advance open, Ethernet -based AI infrastructurefor scale-up,” said Hock E. Tan, president and CEO at Broadcom.“Our highperformance silicondelivers industry-leading ultra-low latency, massive performance, and loss-less networkingwiththe scalability and efficiency modern AI workloads require. Together with HPE and AMD, we are enabling customers to build powerful AI data centers with standard Ethernet, maximizingchoice and flexibility while deliveri ng exceptional scale.”
“For more than a decade, HPE and AMD have pushed the boundariesof supercomputing, delivering multiple exascale-class systems and championing open standards that accelerate innovation,” said Antonio Neri, president and CEO at HPE. “With the introduction of the new AMD ‘Helios’ and our purpose-built HPE scale-up networking solution, we are providing our cloud service provider customers with faster deployments, greater flexibility, and reduced risk in how they scale AI computing in their businesses.”
“HPE has been an exceptional long-term partner to AMD, working with us to redefine what is possible in high-performance computing,” said Dr. Lisa Su, chair and CEO at AMD. “With ‘Helios,’ we’re taking that collaboration further, bringing together the full stack of AMD compute technologies and HPE’s system innovation to deliver an open, rack -scale AI platform that drives new levels of efficiency, scalability, and breakthrough performance for our customers in the AI era.”
