Which scenarios suit A100, H100, B100 and B300 GPUs?

A100: mature, for large-model training/fine-tuning and existing projects. H100: high compute for very large model training, with clear FP8 advantages. B300: the mainstream choice in China, balancing training and inference with strong value. B100: next-generation high-end card for very large compute clusters.