Nvidia broke its speed record thanks to a new AI chip

Yesterday, Nvidia announced that its upcoming H100 GPU “Hopper” Tensor Core installed new ones

performance records during hisdebuted in industry standard benchmark MLPerf, posting results 4.5 times faster than the A100, which is currently Nvidia's fastest production AI chip. Ars Technica reports this. Discuss

MPerf benchmarks measure workloads"outputs" that demonstrate how well the chip can apply a pre-trained machine learning model to new data. A group of industry companies known as MLCommons developed the MLPerf benchmarks in 2018 to provide a standardized metric for presenting machine learning performance to potential clients.

In particular, the H100 performed well inBERT-Large benchmark, which measures the performance of natural language processing using the BERT model developed by Google. Nvidia attributes this particular result to the Transformer Engine of the Hopper architecture, which specifically speeds up the training of transformation models. This means that the H100 can accelerate future natural language models like OpenAI's GPT-3, which can compose writing in a variety of styles and chat conversations.

The chip, which is still in development, is predicted to replace the A100 as the company's flagship data center GPU.