NVIDIA/cccl

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

2488 stars · 476 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

CUDA Core Compute Libraries (CCCL) is NVIDIA's unified repository for Thrust, CUB, and libcudacxx, providing high-performance C++ abstractions for CUDA GPU programming. It serves as the standard library for CUDA C++, offering parallel algorithms, cooperative kernel building blocks, and host/device-compatible standard library implementations.

Use cases

• write parallel algorithms on GPUs with a high-level C++ API
• build custom CUDA kernels with block-wide and warp-wide cooperative primitives
• use the C++ standard library in both host and device code
• achieve performance portability across GPUs and multicore CPUs
• reduce, scan, sort, and transform large arrays on NVIDIA GPUs
• write safe and efficient CUDA C++ without hand-tuning every kernel

When to avoid

• you are targeting non-NVIDIA GPUs or CPU-only parallelism without a supported backend
• you need a runtime or driver-level API rather than C++ header libraries
• your project does not use C++ or CUDA

Facets

library · maturity active

gpu-computing concurrency math developer-tools gpu-computing developer-tools performance programming-languages cpp cross-platform cuda thrust cub libcudacxx parallel-algorithms hpc nvidia algorithms gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NVIDIA/cccl")