A k-means variant that is up to 7x faster than FAISS and Scikit-Learn on CPUs and 4x faster than cuVS on GPUs.
March 23, 2026
Original Paper
A Super Fast K-means for Indexing Vector Embeddings
arXiv · 2603.20009
The Takeaway
Provides a massive speedup for a core ML primitive (vector indexing) by pruning irrelevant dimensions and introducing an 'Early Termination by Recall' mechanism. This is immediately actionable for anyone managing large-scale vector databases.
From the abstract
We present SuperKMeans: a k-means variant designed for clustering collections of high-dimensional vector embeddings. SuperKMeans' clustering is up to 7x faster than FAISS and Scikit-Learn on modern CPUs and up to 4x faster than cuVS on GPUs (Figure 1), while maintaining the quality of the resulting centroids for vector similarity search tasks. SuperKMeans acceleration comes from reducing data-access and compute overhead by reliably and efficiently pruning dimensions that are not needed to assign