AI Term of the Moment
Apple HomeView
Definition: GPU rack
A cabinet containing multiple servers designed for data center operation. For AI data centers, the servers contain a combination of CPUs and GPUs. For example, in 2026, NVIDIA introduced its Vera Rubin rack with 36 Vera CPUs and 72 Rubin GPUs, and AMD introduced its Helios rack with 18 EPYC CPUs and 72 Instinct GPUs. Although GPU quantity is the same, AMD's EPYC CPU has more processing cores than the Vera CPU from NVIDIA.
AMD Agentic AI Rack
AMD's Helion server racks were designed for agentic AI workloads, which require extensive CPU processing rather than GPU calculations (see AI agent). Introduced in 2026, Helion racks are powered by AMD's EPYC CPUs totaling 32 thousand cores for agentic workloads. Other Helion racks have a combination of CPUs and AMD Instinct GPUs for AI training and inference. See AI agent. (Image courtesy of AMD.)