NVIDIA Launches ‘Rubin CPX’ AI GPUs with 128 GB GDDR7 Memory for High-Value Inference Tasks

NVIDIA has made headlines with the introduction of a groundbreaking series of AI GPUs known as the Rubin CPX AI chip, designed to deliver exceptional inferencing capabilities when utilized in a rack-scale cluster configuration.

NVIDIA’s Rubin CPX GPU: A New Benchmark in Rack-Scale AI Performance

Recognizing the growing importance of AI inferencing in computational advancements, NVIDIA has embarked on a new journey with its ‘CFX’ lineup. The inaugural product, the Rubin series, was showcased during the AI Infra Summit. Positioned primarily for long-context AI applications, the Rubin CPX GPU is set to complement the existing Rubin GPUs and Vera CPUs, heralding what NVIDIA describes as a “revolution”in the efficiency of AI inference.

The Rubin CPX boasts impressive specifications, including 30 petaFLOPs of NVFP4 compute power and 128 GB of cutting-edge GDDR7 memory. It will be integrated into the specially designed NVIDIA Vera Rubin NVL144 CPX rack, which will house 144 Rubin CPX GPUs, 144 Rubin GPUs, and 36 Vera CPUs, collectively achieving a staggering eight exaFLOPs of NVFP4 compute. This marks a significant 7.5-fold increase over the Blackwell Ultra system and aims to handle one million-token context AI inference workloads with enhanced performance through innovations like Spectrum-X Ethernet.

Vera Rubin NVL144 CPX compute tray with labels Rubin, Vera, Rubin CPX, ConnectX-9 on a black background.

This platform is projected to deliver a remarkable “30x to 50x return on investment, ”positioning the Vera Rubin NVL144 CPX rack as a vital tool for overcoming the limitations currently faced in developing next-generation generative AI applications. While more configurations of the Rubin CPX are expected, specifics remain undisclosed. However, its integration of GDDR7 memory instead of HBM suggests a more cost-effective solution for many users.

NVIDIA is adeptly navigating the complexities of the AI landscape, effectively minimizing the chances for competitors to gain an edge. The imminent launch of the next-gen Rubin AI lineup next year promises to elevate compute capabilities to unprecedented heights.

Source & Images