# E-means researchers report faster clustering on GPUs and CPUs

_Published Monday, November 9, 2026 at 2:42 PM EST · Science, AI · Latest · Tier 2 — Notable_

![E-means researchers report faster clustering on GPUs and CPUs — Primary](https://e-means.github.io/assets/emeans_vs_lloyd_readme_v9.png)

The e-means project reports a 10-45× reduction in GPU runtime against Flash-KMeans and an 8-15× reduction on CPUs against Faiss. Its clustering method reaches a test loss below Lloyd's algorithm at 25 epochs in 1.2 to 3 epochs, according to the project's results.

The method groups similar data points using small-batch updates that gradually discount older assignments. Scaling laws automatically set batch and step sizes, while splitting active clusters revives empty ones. Specialized kernels translate the algorithm into runtime gains.

The researchers validated e-means on 24 embedding and synthetic datasets and configurations. The method targets static datasets, rather than continuously changing distributions such as those used in codebook learning.

## Sources

- [e-means.github.io](https://e-means.github.io/)

---
Canonical: https://techandbusiness.org/newswire/gqeU547-Yq-yHNfUwD98hq
Published: 2026-11-09T19:42:50.046Z
Story chronology: 2026-11-09T18:45:43.000Z
Retrieved: 2026-11-09T21:47:51.628Z
Publisher: Tech & Business (techandbusiness.org)