"Frontier Supercomputer Trains ChatGPT-Like Model Using Just 8% of Its GPUs"
The Frontier supercomputer, powered by AMD's EPYC CPUs and Instinct MI250X GPUs, has set a new industry benchmark by training a 1 trillion parameter language model, showcasing the potential of its 37,000 GPUs. This achievement, detailed in an Arxiv submission report, demonstrates Frontier's capabilities in large language model (LLM) training and its status as the world's leading and second most efficient supercomputer. With only 3,000 of its GPUs used for this task, the results hint at the immense potential for further scaling and advancements in AI and HPC performance.