NVIDIA Collaborates with Meta and Oracle to Enhance AI Infrastructures

Implementation in Network Infrastructures

Meta and Oracle have decided to implement NVIDIA's Spectrum-X Ethernet switches in their network infrastructures, aiming to boost the efficiency of AI training and reduce analysis times. Meta plans to introduce these switches into its Facebook Open Switching System for optimized management of large-scale networks. Meanwhile, Oracle intends to build giga-scale AI supercomputers, using these technologies to interconnect millions of GPUs with increased efficiency. This collaboration highlights both companies' commitment to standardize on NVIDIA's solutions to foster the next generation of AI infrastructures.

Strengthening Cloud AI Infrastructure

Oracle relies on its partnership with NVIDIA to strengthen its cloud infrastructure dedicated to artificial intelligence by interconnecting an unprecedented number of GPUs, which should enhance AI training and rapid deployment of AI applications. According to Oracle, this strategy will allow clients to more quickly benefit from the latest advancements in generative AI and reasoning. Concurrently, Meta aims to integrate the Spectrum-X Ethernet solutions into its Minipack3N switch and the FBOSS system, reinforcing its open network approach and optimizing its infrastructure to support increasingly large AI models.

Capabilities of the Spectrum-X Platform

NVIDIA's Spectrum-X Ethernet platform is designed to meet the demands of AI models containing trillions of parameters, with a particular focus on efficiency and scalability at large scales. This system stands out by offering a data throughput of 95%, thus enhancing performance compared to standard commercial Ethernet solutions, which are often limited by flow collisions. This development could influence data center strategies worldwide, enabling an efficient link between geographically distributed infrastructures and supporting the construction of vast AI factories.