NVIDIA has launched the NVIDIA DGX SuperPOD, based on the Rubin platform, each containing 600TB of RAM.

Following NVIDIA's unveiling of the Rubin architecture details today, they also introduced the NVIDIA DGX SuperPOD, a server cluster deployment model for each chip variant.

Two versions of the NVIDIA DGX SuperPOD for Rubin are available: a DGX Vera Rubin NVL72 rackmount version using the Vera ARM architecture CPUs, and a DGX Rubin NVL8 rackmount version using the x86 architecture.

The DGX Vera Rubin NVL72 rackmount SuperPOD features 576 Rubin GPUs, 600TB of RAM, a total processing power of 28.8 exaflops in FP4 mode, and NVLink connectivity with a maximum data transfer rate of 260TB/s. The DGX Rubin NVL8 SuperPOD incorporates 512 Rubin chips.

Both SuperPOD models are expected to begin shipping in the latter half of this year.