NVIDIA/cub
[ARCHIVED] Cooperative primitives for CUDA C++. See https://github.com/NVIDIA/cccl observed · 2026-08-28
Health v2 · maintenance only
10/100
- Activity 0
- Release rhythm 8
- Longevity 100
Flags: archived
How is this computed?
round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10) — computed 2026-09-03. Adoption (stars, forks) is never an input.
- gap_med: n/a
- age_days: 4947
- days_rel: n/a
- days_push: 1059
- n_releases_24m: 0
Adoption not part of the score
1842 stars · 464 forks observed · 2026-08-28
What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded
CUB is a header-only C++ library providing reusable cooperative primitives (sort, scan, reduction, histogram) for device-wide, block-wide, and warp-wide layers of the CUDA programming model. It is now archived and maintained as part of NVIDIA's unified CCCL repository.
Use cases
- sort large arrays on the GPU with CUDA
- compute parallel prefix scans in CUDA kernels
- block-wide and warp-wide collective reductions
- build GPU histograms efficiently
- reuse high-performance CUDA algorithm primitives in custom kernels
When to choose
- you need battle-tested parallel primitives for CUDA C++ kernels
- you want device-, block-, or warp-level algorithms without writing them yourself
- you are already using the CUDA Toolkit or NVIDIA HPC SDK where CUB is bundled
When to avoid
- you are starting a new project - use the successor nvidia/cccl (CUDA C++ Core Libraries) instead
- you need CPU-only parallel algorithms
- you need a runtime library rather than header-only templates
Facets
library · maturity maintenance
concurrency developer-tools gpu-computing developer-tools cpp cuda gpu-computing parallel-algorithms header-only archived nvidia algorithms gpu
1 source
- readme: https://github.com/NVIDIA/cub · fetched 2026-08-28 · 148caebc08a5
Member repositories
| Repository | Role | Health v2 |
|---|---|---|
| NVIDIA/cub | main | 10 |
For agents
Data as of 2026-08-30T08:39:29.467469+00:00 · Report a problem