Nvidia publishes first Blackwell B200 MLPerf results: Up to 4X faster than its H100 predecessor, when using FP4

Nvidia’s Blackwell can deliver 3.7X – 4X higher performance than Hopper H100 processors in generative AI inferencing, according to Nvidia. That’s using FP4 vs. FP8 and comparing a single B200 to four H100 GPUs as well. 

Nvidia’s Blackwell can deliver 3.7X – 4X higher performance than Hopper H100 processors in generative AI inferencing, according to Nvidia. That’s using FP4 vs. FP8 and comparing a single B200 to four H100 GPUs as well.