Tesla CEO and X owner Elon Musk has swiftly promoted the capabilities of xAI and Grok across its social handles. To improve the AI business, the billionaire reportedly told investors that xAI is considering plans to build a supercomputer that may power the upcoming iterations of the Grok chat model, according to The Information.
Citing a presentation for investors, the news outlet added that Musk plans to make the proposed project running by the fall of next year. In this initiative, the company may partner with Oracle to build the technology. Once the supercomputer materialises, the group of flagship Nvidia H100 GPU chips may turn out to be 4x the size of the biggest GPU cluster, which currently exists in the world.
The H100 series GPUs presently dominate the AI-backed data center landscape and can be tough to obtain due to their rising demands. Meanwhile, Musk launched xAI as an AI rival initiative to ChatGPT maker OpenAI and Google. The search leader has been betting big on AI across its diverse product line and grabbing varied responses.
Previously, Musk stated Grok 2 was trained via 20,000 Nvidia H100 GPUs and the chat model's next iteration will require over 1,00,000 H100 chips. Meanwhile, xAI recently introduced Grok 1.5 Vision Preview, its first-generation multimodal model which can process visual information such as docs, diagrams, charts, screenshots, photos and more.
Grok-1.5V will be available for early testers and existing users soon. Moreover, the company's Grok-1.5 debuted as an updated iteration of Grok's previous version. According to xAI, it gets improved reasoning and a context window of 1,28,000 tokens.
Also Read: Grok-1.5 AI Will Be Available Next Week, Grok 2 AI Now In Training, Confirms Elon Musk
It will be available for eligible X (formerly called Twitter) users and early testers soon. "One of the most notable improvements in Grok-1.5 is its performance in coding and math-related tasks," xAI said. The memory capacity of the version is 16 times more than the previous context length.