The MLPerf benchmark, which tests computer systems’ ability to train machine learning neural networks, has entered the generative AI era. Companies like Intel, Nvidia, and Google competed in training large language models (LLM), including GPT-3, using massive systems. Nvidia’s 10,000 GPU supercomputer was the largest ever tested. Overall, there was a 2.8-fold performance boost in the past five months and a 49-fold boost since MLPerf began five years ago. Nvidia continued to dominate the benchmarks with its H100 GPUs, while Microsoft’s Azure was just seconds behind. Intel made progress with its Gaudi 2 accelerator chip, showcasing lower training times.
https://spectrum.ieee.org/generative-ai-training