
10–45× fasterH100 clustering vs k-means · authors’ own tests
e-means clusters similar data 10–45× faster on GPUs
e-means, a clustering method from Meta FAIR and École Nationale des Ponts et Chaussées researchers, groups similar data vectors. Their arXiv preprint reports 10–45× faster GPU runtimes than established k-means implementations when reaching the same clustering error.
Sources
Papere-means: outperforming k-means (in one-tenth the epochs)Codefacebookresearch/faiss: A library for efficient similarity search and clustering of dense vectors.Written by AI from the sources