NVIDIA Brings Multi-GPU UMAP, Processing 870GB in 8 Minutes

NVIDIA added multi-GPU support for UMAP in cuML and cuVS, using balanced clustering partitions and merged local kNN graphs. It processed an 870GB MIRACL dataset in just 8 minutes.

2026-08-19 ~ 2026-08-20 · 2 related posts

• NVIDIA publishes multi-GPU method for massive-scale UMAP in minutes — leland_mcinnes · 2026-08-19
• NVIDIA Multi-GPU UMAP Processes 870GB of Data in Just 8 Minutes — leland_mcinnes · 2026-08-20