After getting booted from OpenAI and left with “bad taste”, Elon Musk is no stranger to the AI trend even after getting booted from OpenAI and left with a “bad taste”. With Tesla and formerly Twitter now X having a wonky and quite the unstable financial future, there’s no doubt that he wants to be back in the AI (and most likely Gen-AI) game through xAI. And now we have confirmed the news of who will be supplying the hardware for training.

Dell and Supermicro

Citing Reuter’s report, Dell Technologies and Supermicro are taking on the task with Dell assembling half of the racks for the supercomputer required for the upcoming iteration of Grok. As for Supermicro, they are known for having a super close relationship with NVIDIA and have the “legendary” story of investing in R&D of liquid-cooling technology for server racks despite no one in the industry thought it was a great idea at that time. And now, it is clear that in selected cases (and growing more moving forward), the cost-performance ratio of direct liquid cooling will win over air-cooling solutions.

Back to xAI, with Grok 2 estimating about 20K NVIDIA H100 GPUs for efficient training while Grok 3 bumps that number up to 100K, Elon says he wants the system up and running by fall 2025. Just some thoughts though, will he still stick with Hopper GPUs when Blackwell rolls around? I mean the “Buy More Save More” meme has been around for years. Won’t Jensen convince him to get the latest and greatest? I mean he is the one who hand-delivered the very first DGX-1 system to Elon containing 8x Tesla-based P100 GPUs. Pretty sure there will be more room for “discussions”, just without the public staring at them.


Related Posts

Subscribe via Email

Enter your email address to subscribe to Tech-Critter and receive notifications of new posts by email.

Leave a Reply