A colossal AI supercomputer, comprising 300 racks and boasting nearly 10,000 Nvidia AI GPUs, was listed for sale on eBay for $60 million. The system, offered by IT parts vendor Core 4 Solutions, also featured 4,200 AMD EPYC CPUs and a substantial 1.4 petabytes of memory. The listing, which concluded on October 28, even included free shipping for U.S. customers via FedEx Ground or Home Delivery.
Each rack within the cluster was equipped with four compute nodes, each containing eight Nvidia A100 40GB GPUs, eight Nvidia Connect network interface cards, and two AMD EPYC 64-core CPUs. Additionally, every rack included 1 TB of DDR4 memory and three 1U management nodes, each also fitted with two EPYC CPUs. This configuration amounts to a total of 9,600 Nvidia A100 GPUs and 4,200 AMD EPYC CPUs across the entire cluster.
The Nvidia A100 GPU, built on the Ampere architecture, is an enterprise-grade processor designed for AI workloads and data analytics, not for gaming. Each A100 GPU features 6912 CUDA cores. The sheer volume of memory available in the cluster, 1.4 petabytes, is enough to supply approximately 90,000 gaming PCs with 16GB of RAM each.
The listing for the $59.7 million cluster has now ended, and it remains unclear whether the system was sold. The significant demand for high-end computing components, particularly GPUs and memory, driven by the boom in AI development, is reflected in the scale and cost of such hardware. Companies in the AI sector are investing heavily in infrastructure, contributing to the high prices and demand for components like those found in this massive rack system.