Project Rainier: The New Summit of AI Designed by AWS | Gpu hardware list | Cpu hardware list and their functions | Intel hardware | Turtles AI

AWS is building Project Rainier, a massive EC2 UltraCluster cluster with over 400,000 Trainium2 chips distributed across multiple US data centers. Designed to train frontier AI models like Claude with exaflop compute power and high-efficiency networking.

Key Points:

• EC2 UltraCluster cluster with tens of thousands of UltraServers, each with 64 Trainium2 chips connected via NeuronLink.
• Five times the compute capacity of Anthropic’s current largest cluster.
• Distributed architecture across multiple data centers, featuring NeuronLink networking and Elastic Fabric Adapter (EFA) for ultra-low latency.
• Trainium2 delivers up to 4× the performance of the previous generation and 30-40% better price/performance than AWS P5 GPUs.

AWS designed Project Rainier as a modular, distributed architecture, with UltraServers integrating four Trn2 instances into a single physical node: 64 Trainium2 chips connected via NeuronLink, delivering up to 332 petaflops per node in sparse FP8 operations. The decision to distribute the cluster across multiple sites raises engineering complexity, but optimizes access to power, cooling, and scalability.

Anthropic, thanks to an $8 billion AWS investment, will use Rainier to train its Claude models — with a five-fold increase in computing power compared to its existing cluster. The adoption of Trainium2 — already tested internally by Apple — aims to reduce dependence on Nvidia, offering competitive performance at lower costs.

The AWS technology roadmap includes Trainium3, coming in 2025: it promises a doubling of performance and 40-50% increased energy efficiency. This is part of a vertical integration strategy, where AWS controls hardware (chips and servers), networking (NeuronLink, EFA), and data centers, to optimize costs, performance, and sustainability.

Thanks to proprietary designs for power, cooling, and building materials, the new data centers reduce mechanical energy by up to 46% and water use per kWh drops to 0.15L, less than half the industry average.

Project Rainier represents a paradigm on which AWS aims to build exaflop-scale compute capacity, simplifying complex operations and promoting efficiency and technology independence.

An emerging platform that confirms how large-scale AI computing is becoming critical infrastructure for the future of AI.