Datacenter GPUs are built for 24/7 operation in cooled server environments. They differ from consumer cards in several ways: passive cooling (depending on chassis airflow rather than onboard fans), ECC memory, HBM-based VRAM, higher VRAM capacities (40-200+ GB), and support for high-bandwidth interconnects (NVLink, Infinity Fabric).
Examples include NVIDIA's H100, H200, B200, A100, and L40S; AMD's MI300X and MI325X; Intel's Gaudi 3. Pricing for datacenter parts is substantially higher than equivalent-spec consumer cards, reflecting both higher manufacturing costs and the enterprise-support market.
AIMC's GPU canonicalization classifies cards into three categories — datacenter, workstation, consumer — based on form factor, memory type, and intended use. The category drives both the fit-score weighting and the default sort order on /gpu category pages.