SoylentNews

upstart [soylentnews.org] writes:

████ # This file was generated bot-o-matically! Edit at your own risk. ████

Musk plans largest-ever supercomputer, report says - Taipei Times [taipeitimes.com]:

Musk plans largest-ever supercomputer, report says AI COMPETITION: The planned supercomputer would be ‘at least four times the size of the biggest GPU clusters that exist today,’ Elon Musk was quoted as saying

• AFP, WASHINGTON

Billionaire tech mogul Elon Musk has told investors he plans to build a supercomputer dubbed “gigafactory of compute” to support the development of his artificial intelligence (AI) start-up xAI, an industry news outlet reported on Saturday.

The Tesla Inc CEO wants the supercomputer — which would string together 100,000 Nvidia H100 graphics processing units (GPU) — to be operational by fall next year, and he “will hold himself personally responsible for delivering it on time,” The Information said.

Nvidia’s H100 family of powerful GPUs dominate the data center chip market for AI, but can be hard to obtain due to high demand.

The xAI logo is displayed on a smartphone screen in New York on Dec. 7 last year.

Photo: Bloomberg

The planned supercomputer would be “at least four times the size of the biggest GPU clusters that exist today,” such as those used by Meta Platforms Inc to train its AI models, Musk was quoted as saying during a presentation to investors this month.

Since OpenAI’s generative AI tool ChatGPT exploded on the scene in 2022, the technology has been an area of fierce competition between tech giants Microsoft Corp and Google Inc, as well as Meta and start-ups like Anthropic and Stability AI Inc.

Musk is one of the world’s few investors with deep enough pockets to compete with OpenAI, Google or Meta on AI.

His company xAI is developing a chatbot named Grok, which can access social media platform X, also owned by Musk, in real time.

Earlier this year, Musk said training the Grok 2 model took about 20,000 Nvidia H100 GPUs, adding that the Grok 3 model and beyond would require 100,000 Nvidia H100 units.

In related news, Tesla shareholders are being urged by a major proxy advisory firm to reject a proposed US$56 billion pay package for Musk, in a blow to the electric-vehicle manufacturer’s board.

Glass Lewis & Co made its recommendation in a report released on Saturday, citing the “excessive size” of the pay deal and the dilutive effect upon exercise.

“Mr. Musk’s slate of extraordinarily time-consuming projects unrelated to the company was well-documented before the 2018 grant, and only expanded with his high-profile purchase of the company now known as X,” Glass Lewis said.

The recommendation to large institutional investors might sway their vote over Musk’s pay at the vehicle manufacturer’s annual meeting on June 13. If the proposal is rejected, the CEO might make good on threats to develop products outside of Tesla.

Next month’s vote would be the second time Musk’s pay package has been put before shareholders. The remuneration deal was originally crafted in 2018, but earlier this year a Delaware judge voided it, saying investors were not fully informed of key details.

If Tesla’s board can show the compensation deal still has broad support, it might help with a legal appeal of the ruling. On the other hand, a loss would be a major embarrassment and demonstrate that investors are losing faith in Musk’s leadership. The vote is only advisory, meaning Tesla can choose to ignore it.

Additional reporting by Bloomberg

;

Original Submission