NVIDIA/cuda-samples v13.4

CUDA Samples v13.4

Repository: NVIDIA/cuda-samples

Tag: v13.4

Published: 2026-09-09T22:14:43Z

Prerelease: no

Release notes: Release 13.4 of the CUDA Samples, supporting CUDA Toolkit 13.4.

Highlights:

• Reorganized the C++ sample CMake build system with shared architecture detection and common build setup.
• Added default build support for sm_103 and sm_107.
• Added new samples for persistent CUDA Python program caching, dma-buf interoperability, and locality domains.
• Modernized several introductory samples, including simpleMultiGPU, simpleCudaGraphs, simpleAtomicIntrinsics, simpleCallback, simpleStreams, clock, vectorAdd, simplePrintf, and simpleAssert.
• Added Windows on Arm build support updates for selected samples.
• Removed obsolete and redundant samples, including several legacy _nvrtc variants and math library samples now maintained in the CUDA Library Samples repository.
• Updated Python samples and test configuration for newer CUDA Python and cuda.core behavior.

See CHANGELOG.md for the full list of changes.