NVIDIA H100 Tensor Core GPUs have been recognized for their exceptional AI performance, particularly in the field of generative AI using large language models (LLMs). These GPUs have set new records in the latest MLPerf training benchmarks for generative AI. In one test, a cluster of 3,584 H100 GPUs completed a GPT-3-based training benchmark in under eleven minutes. NVIDIA’s GPUs have demonstrated high performance across a range of benchmarks, including recommenders, computer vision, medical imaging, and speech recognition. The company’s AI ecosystem is expanding, with submissions from major system makers and more than 30 submissions running on H100 GPUs. NVIDIA’s focus on energy efficiency is also highlighted, with accelerated computing reducing the number of server nodes needed and software optimizations improving performance. The software used for the benchmarks is available for anyone to access.
https://blogs.nvidia.com/blog/2023/06/27/generative-ai-debut-mlperf/