# NVIDIA/nccl-tests

NCCL Tests

Repository: https://github.com/NVIDIA/nccl-tests
Canonical: https://ross.abutalabs.com/products/nccl-tests
Language: Cuda
License: BSD-3-Clause
License Family: permissive
Last push: 2026-08-20T17:45:41+00:00

## Health v2 (maintenance only)
Score: 76/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 98, release rhythm 35, longevity 100
- inputs: {"age_days": 3312, "days_push": 13, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1636, forks 399 (observed 2026-08-28T04:05:14.785482+00:00)

## What it is
NVIDIA's official test suite for NCCL that measures the performance and correctness of collective communication operations across GPUs. It supports single- and multi-node runs via MPI and reports bandwidth metrics like busbw.

## Use cases
- benchmark allreduce bandwidth across multiple GPUs
- verify NCCL collective operations are correct
- measure multi-node GPU communication performance
- profile bus bandwidth for distributed training clusters
- test NCCL performance from 8 bytes to gigabyte message sizes
- validate GPU interconnect performance after cluster setup

## When to choose
- you need to benchmark or validate NCCL collective operations on NVIDIA GPUs
- you are tuning or validating multi-node GPU clusters for distributed training
- you want official, well-maintained NCCL performance baselines

## When to avoid
- you need to benchmark non-NVIDIA GPU communication libraries
- you want a general-purpose GPU compute benchmark rather than communication tests
- your environment has no CUDA/NCCL installation

## Facets
- artifact type: cli-tool
- maturity: active
- function: benchmarking, testing, gpu-computing
- domain: developer-tools, performance, gpu-computing
- platform: cli
- tags: nccl, cuda, mpi, multi-node, collective-communication, hpc, linux

## Member repositories
- NVIDIA/nccl-tests (main) score 76

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:14.785482+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:46:57.452235+00:00, confidence not recorded.
  - readme: https://github.com/NVIDIA/nccl-tests (fetched 2026-08-28T04:05:14.785482+00:00, sha 54b0ea82a608)
- Data as of 2026-08-30T08:39:29.467469+00:00.