Elon Musk’s AI corporate, xAI, not too long ago unveiled a brand new supercomputer dubbed Colossus. And prefer its identify implies, it is large.The pc is a man-made intelligence coaching gadget that Musk says runs on a whopping 100,000 Nvidia H100 chips, a formidable graphics processing unit that is turn out to be vital to the AI race.To position that into standpoint, Meta’s Llama 3 massive language fashion was once educated on 16,000 H100 chips. Meta stated in March it will proceed to put money into its AI infrastructure via including two new 24,000 chip clusters.In different phrases, Musk’s Colossus is robust. And it will assist him catch as much as the AI business’s frontrunners.However some distinguished tech leaders aren’t so certain.LinkedIn cofounder Reid Hoffman informed The Knowledge, a tech newsletter, that the xAI supercomputer was once mere “desk stakes” within the aggressive box of generative AI.Consistent with The Knowledge, Hoffman intended that Colossus handiest permits xAI to catch as much as different, extra complicated AI corporations, like OpenAI and Anthropic.Chris Lattner, the CEO of Modular AI, stated throughout a panel dialogue at The Knowledge’s AI Summit ultimate week that Musk’s heavy reliance on Nvidia’s pricey and finite chips could also be inconsistent with the billionaire’s effort to construct his personal GPU, referred to as Dojo, The Knowledge reported.Meta, Microsoft, Alphabet, and Amazon are all creating their very own AI chips whilst they proceed to stockpile Nvidia GPUs.”The adaptation is that Elon has been running on Dojo for a few years now,” Lattner informed Industry Insider in an electronic mail.
Musk has expressed worry in regards to the demanding situations of obtaining extra of Nvidia’s extremely sought-after chips and stated that his Dojo undertaking will assist lower his corporate’s dependence at the chipmaker.”We do see a trail to being aggressive with Nvidia with Dojo,” Musk stated throughout a Tesla profits name in July. “We more or less don’t have any selection.”When speaking about Colossus on X in early September, Musk stated he targets to double the scale of the supercomputer to 200,000 chips in a couple of months.He stated the cluster was once in-built simply 122 days — an outstanding feat that no different corporate has matched, in keeping with The Knowledge.It is unclear if Colossus runs 100,000 GPUs on the identical time, which will require subtle networking era and numerous power.”Musk prior to now stated the 100,000-chip cluster was once up and operating in past due June,” The Knowledge reported. “However at the moment, an area electrical application stated publicly that xAI handiest had get admission to to a couple of megawatts of energy from the native grid.”Ultimate month, CNBC reported that an environmental advocacy crew had complained that xAI was once operating fuel generators to supply extra energy for the information middle with out authorization.The hole reported that the Southern Environmental Legislation Heart wrote in a letter to the native well being division that xAI had put in and was once running a minimum of 18 unpermitted generators “with extra probably at the manner” to complement its large power wishes.The native application, Memphis Mild, Gasoline and Water, informed CNBC it has equipped 50 megawatts of energy to xAI because the starting of August, however that the ability calls for an extra 100 megawatts to function.Information cluster builders informed The Knowledge that this may handiest energy a couple of thousand GPUs. Musk’s corporate would want any other electrical substation to get sufficient energy to run 100,000 chips.Hoffman and Musk didn’t right away reply to requests for remark from Industry Insider.
