xAI is using Nvidia Ethernet Networking for Grok’s Supercomputer AI cluster

Nvidia has announced that the new xAI Supercomputer cluster is using the Spectrum-X Ethernet networking platform. Both firms have partnered for generative AI and large language model (LLM) technologies.

This AI cluster is located at xAI’s facility in Memphis, Tennessee, and houses 100,000 Nvidia Hopper Tensor Core GPUs to train the latest Grok 2 LLMs. This model family has two versions – Grok 2 mini and Grok 2, the AI company has packaged these two models based on their names.

Two of these models bring generative image features and performance for fast replies and vision capabilities to analyze pictures and screenshots.

Nvidia xAI

Spectrum-X

This ethernet platform is designed to improve the performance and efficiency of Ethernet-based AI clouds. It can provide up to 1.6 times greater AI networking in multi-tenant environments.

It combines a Nvidia Spectrum-4 Ethernet switch and BlueField 3 SuperNIC. Its optimizations can reduce the runtime of large generative AI models and fast data delivery.

It is power efficient for the AI environment and reduces power consumption to lower operational costs. Furthermore, Spectrum-X provides high scalability to 256 200G ports in a single hop or 16,000 ports in a two-tier leaf/spine topology.

(source)