SuperPOD · DGX SuperPOD
NVIDIA's reference cluster design wiring many DGX systems into a validated, scalable AI supercomputer.
Current numbers
250 / 124 GB/sB300 SuperPOD, Standard 80/40 GB/s per SU; Enhanced read/write per SU (512-GPU Ethernet RA or 576-GPU XDR RA)
280 / 140 GB/sDGX SuperPOD with Vera Rubin NVL72, Standard 90/45 GB/s per SU; Enhanced read/write per 8-system (576-GPU) SU
576 GPUs / SUNVIDIA GB200 SuperPOD reference SU, separate from the 512-GPU teaching case