xAI to boost supercomputer cluster with 200k Nvidia H100 GPUs

Nvidia has confirmed that xAI will increase up to 200k Supercomputer AI cluster GPUs for the next phase of the Grok large language model (LLM).

The AI chip maker has details about its partnership with xAI revealing that the current Supercomputer AI cluster is equipped with 100,000 H100 GPUs at its facility in Memphis, Tennessee.

These GPUs are used to train the latest Grok 2 and Grok 2 Mini model family. The two can provide fast performance and bring generative AI image capabilities.

Also, the supercomputer is featured with Nvidia’s Spectrum-X Ethernet networking platform. Aside from the initial capacity, xAI has plans to double the H100 capacity to 200,000 soon.

Nvidia said that xAI built its supercomputer AI cluster named “Colossus” in 122 days. In comparison, other companies would have taken more than a year to complete this task. With its new network architecture, the Colossus has achieved fast network performance.

“xAI has built the world’s largest, most powerful supercomputer,” said a spokesperson for xAI. “NVIDIA’s Hopper GPUs and Spectrum-X allow us to push the boundaries of training AI models at a massive scale, creating a super-accelerated and optimized AI factory based on the Ethernet standard.”

H100s are the first choice in the AI industry for easy data center integration, scalability, and powerful generative AI processing capabilities.

In August, xAI released the Grok 2 model family for X Premium and Premium+ subscribers. Looking at the data, Grok 2 has surpassed Grok 1.5 with big numbers on all of the benchmarks. It also comes close to GPT-4 and Anthropic’s Claude 3 Opus.

Next, the AI company plans to improve AI technologies with Grok version 3, the AI firm to further improve performance and efficiency. This new model is expected to launch later this year and we may see the additional cluster capacity going online prior to this release.

(source)