Enhanced GH200 Super Chip: Next-Gen AI Power

NVIDIA has just introduced the upgraded GH200 super chip, a groundbreaking GPU featuring the latest HBM3E technology. This powerful innovation is set to hit production in the second quarter of the upcoming year.

The GH200 takes advantage of the Hopper architecture, allowing seamless integration of next-generation memory without the need for a complete GPU redesign. This showcases the remarkable versatility of the Hopper framework.

NVIDIA underscores the growing need for cutting-edge techniques in AI server data training to deploy advanced global technologies. Even the best individual GPUs can't fully meet these demands, leading to the development of super GPUs by linking the right number of GPUs together, thereby offering an interactive AI experience.

The GH200 model boasts a 72-core GraceGPU and an impressive 141GB of HBM3E memory. HBM3E not only enhances capacity but also elevates speed, delivering an exceptional 5TB per second bandwidth. Compared to the existing H100, each GPU's capacity has surged by 1.7 times, while the bandwidth has leaped by an impressive 1.55 times.