NVIDIA Blackwell B300 GPUs are part of the Blackwell Ultra generation and are designed for large-scale AI training, inference, and reasoning. Compared with the original Blackwell generation, they offer up to 288 GB of HBM3e memory per GPU, 50% more NVFP4 computing performance, and up to twice the attention-layer performance. This makes them particularly suitable for large models, long context windows, high-volume inference, and workloads serving many users simultaneously.