PyTorch nested_graph_breaks switch delivers 4.53x speedup on NVIDIA B200

Developer cfregly benchmarked a 64-ATen-op graph on NVIDIA B200 with PyTorch 2.14.1 (CUDA 13.0), finding that simply enabling the new nestedgraphbreaks=True switch yields a 4.53x steady-state speedup for the same computation.

2026-10-26 ~ 2026-10-26 · 2 related posts

• PyTorch nested graph breaks deliver 4.53× steady-state speedup on B200 — cfregly · 2026-10-26
• Torch.compile Flag Cut 16 Regions to 2, Delivering 4.53x Speedup on B200 — cfregly · 2026-10-26