gunrock/gunrock

Health v2 · maintenance only

Adoption not part of the score

1095 stars · 228 forks observed · 2026-08-28

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

Gunrock is a CUDA/C++ library for GPU-accelerated graph analytics built on a high-level, data-centric abstraction over vertex and edge frontiers. It lets developers implement graph primitives that scale from one to many GPUs with minimal GPU programming knowledge, and now also supports AMD GPUs via ROCm/HIP.

Use cases

• run graph algorithms like BFS, PageRank, and SSSP on GPUs
• implement custom graph primitives with a high-level GPU programming model
• accelerate graph neural network workloads on NVIDIA or AMD GPUs
• process large sparse graphs in parallel across multiple GPUs
• prototype graph analytics without writing low-level CUDA kernels

When to choose

• you need high-performance graph analytics on GPU hardware
• you want to develop new graph primitives quickly with minimal CUDA expertise
• you need multi-GPU scaling within a node
• you want a library that supports both NVIDIA CUDA and AMD ROCm backends

When to avoid

• you only need CPU-based graph processing
• your graphs are small enough that GPU overhead outweighs gains
• you need a turnkey graph database rather than a programmable analytics library
• you cannot use C++17 or a modern CUDA/ROCm toolchain

Facets

library · maturity active

graphics math machine-learning developer-tools graph-processing gpu-computing machine-learning cpp windows cuda graph-analytics gpu-computing graph-algorithms parallel-computing graph-neural-networks sparse-matrix hpc rocm hip high-performance-computing algorithms gpu linux macos

Member repositories

For agents

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