(Image credit score: Getty Photos / Carina Johansen)
We today bought a look for of what $1 billion worth of AI GPUs looks love when Elon Musk shared a temporary video tour of Cortex, X’s AI coaching supercomputer currently under constructing at Tesla’s Giga Texas plant.
More today, Musk took to his social media platform to screech that Colossus, a brand original 100k H100 coaching cluster, is now up and working.
Musk claims that Colossus is “potentially the most noteworthy AI coaching scheme on this planet” and that it modified into as soon as built “from start as a lot as place out” in exactly correct 122 days. That’s quite an achievement. Servers for the xAI cluster were reportedly supplied by Dell and Supermicro, with the worth of the venture estimated to be between $3-4 billion.
This weekend, the @xAI team brought our Colossus 100k H100 coaching cluster online. From start as a lot as place out, it modified into as soon as performed in 122 days. Colossus is potentially the most noteworthy AI coaching scheme on this planet. Furthermore, it’s going to double in size to 200k (50k H200s) in about a months. Very just correct…September 2, 2024
The put does Colossus win its title?
Tom’s Hardware notes, “Even if all of these clusters are formally operational and even coaching AI items, it is fully unclear what number of are in actuality online currently. First, it takes some time to debug and optimize the settings of these superclusters. Second, X desires to substantiate that they win adequate energy, and while Elon Musk’s company has been utilizing 14 diesel generators to energy its Memphis supercomputer, they were nonetheless no longer adequate to feed all 100,000 H100 GPUs.”
The Colossus scheme is poised to in a roundabout diagram double in skill, with plans to incorporate an additional 100,000 GPUs – 50,000 H100 items and 50,000 of Nvidia‘s subsequent-gen H200 chips. The supercluster will primarily be historical to coach xAI’s Grok-3, the company’s most contemporary, most superior AI model. We now beget yet to head attempting to search out any mention of storage for the original scheme, but it’s going to will beget to be colossal.
The naming of the original supercomputer has raised more than about a eyebrows, on the opposite hand, with of us noting that it shares its title with a 1970 sci-fi movie (in accordance with a 1966 unique by D.F. Jones) a pair of supercomputer that becomes sentient after being given management of the US nuclear arsenal. Things, predictably, hasten horribly tainted for humanity.
Both the radical and film explore timely topics of AI autonomy, the dangers of relinquishing management to machines, and the ethical implications of artificial intelligence. It’s imaginable that Musk wasn’t attentive to this when the title modified into as soon as chosen for his original AI coaching scheme, and it could need been selected purely to emphasise the sheer scale of the supercluster. On the other hand, with Musk’s song file, it wouldn’t be figuring out if the reference modified into as soon as fully intentional – he is conscious of exactly what he’s doing.