close
close

Elon Musk just launched Colossus – the world’s largest Nvidia GPU supercomputer, built from start to finish in just three months.

Elon Musk just launched Colossus – the world’s largest Nvidia GPU supercomputer, built from start to finish in just three months.

You can say what you want about Elon Musk, but when the technological disruptor sets his mind to something, he plays to win.

His latest AI startup, xAI, was founded just last July and just brought a new supercomputer called Colossus online over Labor Day weekend, designed to train its large language model (LLM) called Grok, a rival to Open AI's better-known GPT-4.

While Grok is limited to paying subscribers of Musk's social media platform X, many Tesla experts speculate that it will ultimately form the artificial intelligence that powers the electric car maker's humanoid robot Optimus.

Musk estimates that this strategic flagship project could ultimately generate annual profits of one trillion dollars for Tesla.

The new xAI data center in Tennessee houses 100,000 Nvidia sets more benchmarks with Hopper H100 processors than any other single AI computing cluster.

“From start to finish, it was done in 122 days,” Musk wrote, calling Colossus “the most powerful AI training system in the world.”

This is not all for xAI either: Musk estimates that he double Colossus' computing capacity could be expanded within months once it can acquire 50,000 of Nvidia's new, more advanced H200 series chips, which are about twice as powerful.

Musk and xAI did not respond to requests for comment. Assets for a comment.

Designed to train Grok-3, the potential next leader in AI models

The pace at which Colossus was built is breathtaking considering xAI only selected its Memphis location in June.

In addition, in the current AI gold rush, several influential technology companies, including Microsoft, Google and Amazon, are competing alongside Musk for Nvidia's coveted Hopper series AI chips.

But the AI ​​entrepreneur is a valued customer of Nvidia and has pledged to spend three to four billion dollars this year on CEO Jensen Huang's hardware – for Tesla alone.

In addition, xAI had a head start because it was able to tap into the supply of AI chips that Tesla had already supplied to the electric car maker.

The Memphis cluster will train Musk's third generation of Groks.

“We hope to release Grok-3 by December, and Grok-3 should be the most powerful AI in the world by that time,” he told conservative podcaster Jordan Peterson in July.

An early beta version of Grok-2 was made available to users just last month.

It was trained on only around 15,000 Nvidia H100s GPUs, but according to competitor chatbot leaderboards, it is already one of the most powerful AI models for large languages ​​by some metrics.

The nearly sevenfold increase in GPU count suggests that Musk has no intention of ceding the race to develop artificial intelligence to OpenAI, which he co-founded in late 2015 after fearing that Google would dominate the technology.

Musk later clashed with CEO Sam Altman and is now suing the company for a second time.

To balance the odds, xAI 6 billion dollars in a Series B funding round in May, with help from venture capitalists such as Andreessen Horowitz and Sequoia Capital, as well as deep-pocketed investors such as Fidelity and the Kingdom Holding of Saudi Prince Alwaleed bin Talal.

Tesla could be the next company to invest in Musk's xAI

Musk has also indicated that he would propose a vote to Tesla's board on whether to invest another $5 billion in xAI, a move welcomed by many shareholders.

But xAI's supercomputer cluster has raised concerns in Memphis because the city government rushed to approve the project, which is designed to bring economic activity back to a part of town that was most recently home to an Electrolux appliance factory.

A major concern is the A burden on the city’s resources. Officials of the municipal utility MLGW estimate that Colossus Up to 1 million gallons of water are required daily to cool the servers, and electricity consumption is up to 150 megawatts.

But Musk is a person who only thinks big, and anything worth doing should be done quickly – otherwise there is a risk of falling behind the competition.

Speaking to Lex Fridman after the podcaster toured xAI's fast-growing operations, Musk said speed is a central part of his five-step management process.

“Everything can be sped up. No matter how fast you think it can be done,” he said last month, “it can be done faster.”

This story originally appeared on Fortune.com

Related Post