B100

The NVIDIA B100 leverages the Blackwell architecture with 192 GB of HBM3e VRAM. It provides datacenter-grade scalability and memory bandwidth, serving as a high-capacity accelerator for organizations building large-scale AI training and inference infrastructure.

Its 192 GB of VRAM supports Large Language Models (LLM) and fine-tuning of frontier-scale models with minimal partitioning overhead. With a benchmark score of 112, the B100 delivers uncompromising performance for enterprise-grade AI workloads including multi-trillion-parameter model training and high-throughput production inference.