Official announcements from xAI or reports from credible tech news outlets

xAI's Liquid-Cooled 'Colossus' Supercomputer in Memphis with 100,000 GPUs, Doubling to 200,000

Oct 28, 2024, 05:53 PM

Elon Musk's artificial intelligence company, xAI, has unveiled Colossus, the world's largest liquid-cooled AI supercomputer, built in partnership with NVIDIA. The supercomputer comprises 100,000 NVIDIA Hopper H100 GPUs and utilizes NVIDIA's Spectrum-X Ethernet networking platform, achieving 95% data throughput and zero latency issues. Colossus was constructed in just 122 days, with initial training beginning after only 19 days when the first rack was powered up. Located in Memphis, Tennessee, the supercomputer powers xAI's Grok AI model. xAI plans to double the size of Colossus by expanding to 200,000 NVIDIA GPUs. "Colossus is the most powerful training system in the world," said Elon Musk. NVIDIA CEO Jensen Huang praised the speed of the installation, highlighting the achievement of xAI in building such a massive supercomputer.