Supports Large Model Training: The GB200 is specifically designed for training and inference of large-scale language models (LLMs), capable of handling models with hundreds of billions of parameters. The NVIDIA DGX GB Rack Scale Systems User Guide is also available as a PDF. Each rack is an NVL72 rack (72-GPU NVL domain). The guide applies to. Ultra-high Computing Power: Compared to its predecessor, the H100, the GB200 offers a 6-fold increase in computing power. When handling multi-modal specific domain tasks, its computing power can reach 30 times that of the H100. These systems utilize both copper and optical interconnects, leading to much discussion in the market about the evolution of “copper” and “optical” technologies. This article focuses on the high-speed interconnect architectures of these. The NVIDIA GB200 functions as a unified high-performance computing system by combining a Grace CPU and two Blackwell GPUs. 8TB/s, which is calculated by bandwidth-oriented individuals in bytes per second (Byte/s).
[PDF Version]