American businessman, head of the social network X, Tesla and SpaceX companies Elon Musk launched the most powerful cluster for artificial intelligence training in the world in a data center in Memphis.

As pointed out by the xAI startup that launched the cluster, the new computing system will be used to train a new version of the GROK artificial intelligence model. This will use NVIDIA H100 GPUs.

“Great work from the xAI team, X team, Nvidia and supporting companies who began training at the Memphis Supercluster at approximately 4:20 a.m. local time. 100,000 liquid-cooled H100 processors on a single RDMA matrix make this the most powerful AI training cluster in the world,” Elon Musk wrote in X.

The GROK 2 chatbot is expected to be released next month, and GROK 3 by December of this year.

According to the source, the new AI cluster of the startup xAI, owned by Elon Musk, surpasses all TOP500 supercomputers in terms of the number of accelerators.

xAI previously leased Nvidia chips for AI training from software developer Oracle. But now the company has decided to build its own data center (DPC). The new data center will surpass in capacity a similar center leased from Oracle.

Each H100 accelerator costs approximately $30,000, and while 20,000 H100 were used to train GROK 2, GROK 3 requires five times the processing power.

ORIENT news