Dell Technologies and Supermicro are joining forces to create an AI supercomputer for Elon Musk’s xAI. Confirmed by CEO Michael Dell on X, the project involves Dell building an “AI factory” with Nvidia GPUs to support xAI’s Grok chatbot.
We’re building a Dell AI factory with @nvidia to power @grok for @xai @elonmusk pic.twitter.com/2aTYLtCBup
— Michael Dell (@MichaelDell) June 19, 2024
Racks from Dell and Supermicro
The server racks for this supercomputer will be divided between Dell Technologies and Supermicro, with each handling half. Musk praised Supermicro’s experience in GPU-accelerated servers, which are essential for AI training and inference. The collaboration between these two companies will be critical to xAI’s endeavors and its Grok AI chatbot.
To be precise, Dell is assembling half of the racks that are going into the supercomputer that xAI is building
— Elon Musk (@elonmusk) June 19, 2024
Massive GPU Implementation
Recently, Elon Musk mentioned that xAI is set to deploy 100,000 Nvidia H100 GPUs, which will be liquid-cooled, in North Dakota. Dell and Supermicro are expected to supply these GPUs. Nonetheless, Musk has voiced concerns about the effectiveness of H100 GPUs after observing Nvidia’s future Blackwell GPUs at the GPU Technology Conference.
Musk speculates that around 300,000 Nvidia B200 GPUs, interconnected by Nvidia’s 800Gb/s ConnectX-8 network cards, could be operational by next summer. The B200 is based on Nvidia’s new Blackwell GPU architecture and features an astounding 208 billion transistors, making it currently the world’s most powerful chip”.
Funding and Disputes
xAI has secured $6 billion from Series B funding to finance a large part of these high-end accelerators. Yet, Musk’s decision to shift 12,000 H100 GPU orders from Tesla to xAI has raised some eyebrows. He defended this choice by explaining that Tesla lacked the necessary infrastructure to use the GPUs effectively, thus reallocating them to xAI seemed preferable.
As xAI focuses on Nvidia GPUs for Grok, Tesla’s future strategy is not as clear. The automaker has incorporated Nvidia GPUs for its self-driving systems but is also working on its proprietary supercomputer, Dojo. Although Musk once allocated over $1 billion for Dojo, recent comments suggest reduced enthusiasm, referring to Dojo as a “long shot” and showing more confidence in orders for Nvidia GPUs.
Last Updated on November 7, 2024 3:48 pm CET
