NVIDIA/cuda-samples
FreeSamples for CUDA Developers which demonstrates features in CUDA Toolkit
FreeFree tier
About NVIDIA/cuda-samples
NVIDIA/cuda-samples is a public GitHub repository containing a collection of sample programs that demonstrate various features of the CUDA Toolkit. The samples cover a range of topics from basics to advanced techniques, written in C++ and Python. The repository is versioned to match CUDA Toolkit 13.3 and includes build support via CMake (version 3.20 or later) for Linux, Windows, and cross-compilation for Tegra devices. It also provides guidance for on-GPU debugging with cuda-gdb. The repository has a large community following with over 9,400 stars and 2,400 forks.
Key Features
Supports CUDA Toolkit 13.3
Includes C++ and Python sample code
Built with CMake (version 3.20 or later)
Cross-compilation support for Tegra devices
On-GPU debugging via cuda-gdb (ENABLE_CUDA_DEBUG flag)
Large collection of sample programs demonstrating CUDA features
Pros & Cons
Pros
- Free and open source
- Comprehensive set of samples covering many CUDA features
- Supports multiple platforms (Linux, Windows, Tegra)
- Regularly updated to match latest CUDA Toolkit
- Large community with many stars and forks
Cons
- Requires NVIDIA GPU and CUDA Toolkit to build and run
- Primarily code examples, not a complete application
- May require familiarity with CMake build system
Best For
Learning CUDA programmingReference implementation for CUDA featuresTesting CUDA Toolkit features across platforms
FAQ
How do I build the CUDA samples?
The samples are built using CMake (version 3.20 or later). On Linux, create a build directory, run cmake .., and then make -j$(nproc). On Windows, use Visual Studio or cmake from the command line.
What CUDA Toolkit version is supported?
The repository currently supports CUDA Toolkit 13.3.
Can I use these samples on Windows?
Yes, the repository provides build instructions for Windows either via Visual Studio or CMake command line using the x64 Native Tools Command Prompt.