Don Kosak

NVIDIA posted their #MLPerf #MachineLearning benchmarks today. Take-away:

2.5X overall speed improvements on A100 (Ampere) due to software improvements since initial release.

6.7X overall speed improvement on new H100 (Hopper) architecture vs. original A100.

Details: https://blogs.nvidia.com/blog/2022/11/09/mlperf-ai-training-hpc-hopper/