AMD Launches Rack-Scale AI Platform Helios, Now in Full Production

At the Advancing AI 2026 event, AMD officially launched the rack-scale AI computing platform Helios. CEO Lisa Su announced that the platform is already in mass production, shipping this quarter with continued scaling next quarter, positioning it as the world's fastest AI rack.

已确认

Helios features a high-density full-rack integration, with a single rack containing 72 Instinct MI455X GPUs, 18 sixth-gen EPYC Venice CPUs, and Pensando DPUs, achieving a total compute power of 2.9 EFLOPS. AMD claims that compared to competitors, Helios delivers 15% more compute, 50% higher HBM4 memory capacity, and 50% greater memory bandwidth. Furthermore, AMD provided an economic metric directly benchmarking against Nvidia: Helios can deliver up to 30% more token output per dollar compared to the Nvidia Vera Rubin NVL72. OpenAI's infrastructure lead also took the stage to discuss their partnership with AMD and these performance figures.

为什么重要

Helios marks AMD's strategic shift from competing solely on chips to system-level delivery, aiming to close the gap with Nvidia in full-rack solutions. Beyond hardware improvements in compute and bandwidth, AMD is aggressively targeting Nvidia's next-generation offerings directly on "token output per dollar" while securing endorsements from major clients like OpenAI, highlighting its aggressive strategy to capture market share in the AI infrastructure space.

2026-07-24 ~ 2026-07-24 · 7 related posts

Full story(4 episodes)→

Primary sources