INSIDE PROMETHEUS : META JOURNEY TO A GIGAWATT - SCALE SUPERCLUSTER

What happened: Meta said the first phase of Prometheus, its gigawatt-scale supercluster in New Albany, Ohio, is complete and was used to train its Muse foundational models, including Muse Wonder. Meta is scaling it toward roughly 1 million GPUs. Key numbers: More than 85% goodput for pre-training and more than 90% for RL runs of Muse Wonder. Why it matters: Meta says the capacity allows larger models and longer context windows than were previously practical, and it is scaling Prometheus with next frontier models from Meta Superintelligence Labs in mind.

Meta said the first phase of Prometheus, its gigawatt-scale supercluster in New Albany, Ohio, is complete and was used to train its Muse foundational models, including Muse Wonder. Meta is scaling it toward roughly 1 million GPUs.

Meta says the capacity allows larger models and longer context windows than were previously practical, and it is scaling Prometheus with next frontier models from Meta Superintelligence Labs in mind.