NVIDIA launches Rubin platform and DGX SuperPOD scale-out

NVIDIA introduced the Rubin platform and positioned Rubin-based NVIDIA DGX SuperPOD (DGX Vera Rubin NVL72 and DGX Rubin NVL8) as the blueprint for large-scale AI deployments, with availability planned in the second half of this year.

• Announcement: NVIDIA introduced the Rubin platform at CES, a codesigned stack of six chips (Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, Spectrum-6 Switch) and announced Rubin-based DGX SuperPOD configurations (DGX Vera Rubin NVL72 and DGX Rubin NVL8) that will be available in the second half of this year. Key technical details include 576 Rubin GPUs per NVL72 SuperPOD (28.8 exaflops FP4 and 600TB memory), 260TB/s aggregate NVLink throughput, and claims of up to 10x reduction in inference token cost versus the previous generation.
• Details & integration: Rubin-based deployments will integrate BlueField-4 DPUs, ConnectX-9 SuperNICs, Quantum-X800 InfiniBand and Spectrum-X Ethernet, and be managed with NVIDIA Mission Control and the NVIDIA AI Enterprise software stack; NVL72 racks combine 36 Vera CPUs, 72 Rubin GPUs and 18 BlueField-4 DPUs per system and NVL8 provides a liquid-cooled x86 on-ramp for Rubin performance.