PCIe (Peripheral Component Interconnect Express) is the standard high-speed bus used to connect GPUs (and other peripherals) to a host system's CPU and memory. Generations have steadily increased bandwidth: PCIe 4.0 x16 delivers ~32 GB/s per direction; PCIe 5.0 x16 doubles that to ~64 GB/s; PCIe 6.0 (emerging) doubles again to ~128 GB/s.
For single-GPU workloads, PCIe bandwidth is rarely the bottleneck. For multi-GPU training, however, PCIe is often the limiting factor unless NVLink (NVIDIA) or Infinity Fabric (AMD) connects the GPUs directly.
PCIe form factor GPUs (H100 PCIe, A100 PCIe, RTX 4090) are easier to deploy in standard server chassis but lose the high-bandwidth multi-GPU interconnect that SXM/OAM forms provide. AIMC's index treats these as distinct canonical SKUs.