Chipmaker Nvidia announced Monday that its Spectrum-X networking technology helped develop startup xAI’s Colossus supercomputer, now recognized as the world’s largest AI training cluster.

Located in Memphis, Tennessee, Colossus serves as the training ground for the third generation of Grok, the xAI suite of large language models developed to power chatbot functionality for X Premium subscribers.

Colossus, completed in just 122 days, began training its first models 19 days after installation. Tech billionaire Elon Musk’s xAI startup plans to double the system’s capacity to 200,000 GPUs, Nvidia said in a statement. statement Monday.

At its core, Colossus is a giant interconnected system of GPUs, each specialized in processing large data sets. When Grok models are trained, they must analyze huge amounts of text, images, and data to improve their responses.

Touted by Musk as the world’s most powerful AI training cluster, Colossus connects 100,000 NVIDIA Hopper GPUs using a unified remote direct memory access network. Nvidia GPU Hoppers handle complex tasks by separating the workload across multiple GPUs and processing it in parallel.

The architecture allows data to move directly between nodes, bypassing the operating system and ensuring low latency and optimal throughput for in-depth AI training tasks.

While traditional Ethernet networks often suffer from congestion and packet loss, limiting throughput to 60%, Spectrum-X achieves 95% throughput without latency degradation.

Spectrum-X makes it easier for large numbers of GPUs to communicate with each other, as traditional networks can get bogged down with too much data.

The technology allows Grok to be trained faster and more accurately, which is essential for creating AI models that respond effectively to human interactions.

Monday’s announcement had little effect on Nvidia stock, which fell slightly. Shares were trading at $141 on Monday, with the company’s market capitalization at $3.45 trillion.

Edited by Sébastien Sinclair