Open-source gpu-computing software
386 open-source libraries, tools and applications tagged gpu-computing, ranked by how closely gpu-computing is their primary purpose, then by adoption and maintenance health. Each entry links to a page with sources, use cases and when to choose or avoid it.
386 products, primary matches first, then adoption-weighted; health v2 shown.
| Product | Health v2 | Stars | Maturity |
|---|---|---|---|
| NVlabs/curobo cuRobo is a CUDA-accelerated Python library for robot motion generation built on PyTorch, CUDA, and Warp. It provides GPU-parallel algorith… | 75 | 1865 | active |
| tile-ai/TileRT TileRT is a tile-based runtime for ultra-low-latency LLM inference, achieving hundreds to over 1000 tokens/s decode speeds on frontier mode… | 75 | 1831 | active |
| GPUOpen-LibrariesAndSDKs/FidelityFX-SDK AMD's FidelityFX SDK (FSR SDK) is a collection of heavily optimized GPU techniques for DirectX 12 and Vulkan applications, including super … | 74 | 1820 | active |
| triple-mu/YOLOv8-TensorRT A library for running YOLOv8 inference accelerated with NVIDIA TensorRT, supporting detection, segmentation, pose estimation, oriented boun… | 72 | 1812 | active |
| chungmin99/pyroki PyRoki is a modular Python toolkit for robot kinematic optimization built on JAX, providing differentiable forward kinematics from URDFs, c… | 50 | 1797 | active |
| moderngpu/moderngpu moderngpu is a header-only C++ productivity library for general-purpose GPU computing built on CUDA. It provides accelerated primitives for… | 47 | 1796 | active |
| 0xSero/turboquant TurboQuant is a Python library implementing near-optimal KV cache quantization for LLM inference, compressing keys to 3-bit and values to 2… | 57 | 1788 | active |
| DTolm/VkFFT VkFFT is an open-source, GPU-accelerated multidimensional Fast Fourier Transform library supporting Vulkan, CUDA, HIP, OpenCL, Level Zero, … | 53 | 1784 | active |
| ILGPU ILGPU is a just-in-time compiler for high-performance GPU programs (kernels) written in .NET languages like C# and F#, entirely in C# with … | 71 | 1762 | active |
| webonnx/wonnx Wonnx is a GPU-accelerated ONNX inference runtime written entirely in Rust, built on wgpu and usable natively or in the browser via WebGPU … | 10 | 1754 | active |
| Zaneham/Booth Booth is an open-source compiler that takes CUDA C, HIP, or Triton kernel source and emits binaries for AMD RDNA 2/3/4 GPUs, NVIDIA PTX, Te… | 79 | 1750 | active |
| OpenNebula/one OpenNebula is an open-source cloud management platform for building and operating private clouds, edge infrastructures, distributed deploym… | 75 | 1748 | stable |
| rusty1s/pytorch_scatter A PyTorch extension library providing highly optimized scatter and segment (sparse update) operations with sum, mean, min, and max reductio… | 57 | 1747 | stable |
| deal.II deal.II is a C++ library for solving partial differential equations using adaptive finite elements, with support for dimension-independent … | 87 | 1737 | stable |
| NVIDIA/FasterTransformer NVIDIA's highly optimized C++/CUDA library for fast inference of Transformer-based models such as BERT, GPT, and encoder-decoder models, wi… | 23 | 6450 | maintenance |
| google-research/circuit_training AlphaChip (circuit_training) is Google's open-source framework for generating chip floorplans using distributed deep reinforcement learning… | 59 | 1730 | active |
| tenstorrent/tt-metal TT-Metal is Tenstorrent's open-source software stack containing TT-NN, a Python and C++ neural network operator library, and TT-Metalium, a… | 99 | 1716 | active |
| pyscf/pyscf PySCF is a Python-based framework for quantum chemistry simulations, providing electronic structure methods such as Hartree-Fock, density f… | 91 | 1699 | stable |
| NCCL NVIDIA's official test suite for NCCL that measures the performance and correctness of collective communication operations across GPUs. It … | 74 | 1664 | active |
| NVIDIAGameWorks/rtx-remix NVIDIA RTX Remix is an open-source platform for remastering classic DirectX 8/9 games with RTX path tracing, DLSS, and AI-upgraded textures… | 75 | 1652 | active |
| psalias2006/gpu-hot GPU Hot is a lightweight, self-hosted web dashboard for real-time NVIDIA GPU monitoring, deployed via a single Docker command. It polls NVM… | 76 | 1633 | active |
| mlcommons/inference Reference implementations of the MLPerf Inference benchmark suite, the industry-standard benchmarks for measuring how fast ML systems run m… | 79 | 1631 | active |
| NVlabs/sionna Sionna is an open-source, GPU-accelerated, differentiable Python library from NVIDIA for research on communication systems. It comprises Si… | 91 | 1615 | active |
| Chainer Chainer is a Python-based deep learning framework that pioneered the define-by-run approach with dynamic computational graphs and automatic… | 23 | 5919 | maintenance |
| Xilinx/brevitas Brevitas is a PyTorch library for neural network quantization supporting both post-training quantization (PTQ) and quantization-aware train… | 93 | 1582 | active |
| RightNow-AI/autokernel AutoKernel is an open-source autoresearch pipeline that takes any PyTorch model, profiles it to find GPU kernel bottlenecks, extracts them … | 44 | 1571 | active |
| utkuozdemir/nvidia_gpu_exporter A Prometheus exporter written in Go that collects NVIDIA GPU metrics by invoking the nvidia-smi binary or reading directly from the NVML dr… | 98 | 1569 | active |
| ByteDance-Seed/Triton-distributed Triton-distributed is a distributed compiler built on OpenAI Triton for computation-communication overlapping on multi-GPU systems. It lets… | 61 | 1559 | active |
| UCCL UCCL is a high-performance GPU communication library written in C++ that provides collectives (as a drop-in NCCL/RCCL replacement), P2P tra… | 70 | 1545 | active |
| intel/llvm Intel's staging area for LLVM upstream contributions and home for Intel LLVM-based projects, most notably the oneAPI DPC++ compiler impleme… | 96 | 1537 | active |
| beehive-lab/TornadoVM TornadoVM is a Java programming framework that JIT-compiles JVM bytecode into GPU kernels targeting NVIDIA CUDA, OpenCL, and Apple Metal at… | 98 | 1523 | active |
| ovg-project/kvcached kvcached is a Python library that brings OS-style virtual memory abstraction to KV cache management for LLM serving and training on shared … | 72 | 1509 | active |
| intel/compute-runtime Intel's open-source Graphics Compute Runtime (NEO) providing OpenCL 3.0 and oneAPI Level Zero compute API support for Intel HD Graphics and… | 98 | 1457 | active |
| NVIDIA/MatX MatX is a C++20 header-only numerical computing library providing NumPy-like tensor expressions for NVIDIA GPUs and multithreaded CPUs. It … | 92 | 1448 | active |
| JuliaGPU/CUDA.jl CUDA.jl is the main Julia package for programming NVIDIA CUDA GPUs, offering a high-level CuArray abstraction, a compiler for writing CUDA … | 99 | 1433 | stable |
| NVIDIA/gdrcopy GDRCopy is a low-latency GPU memory copy library built on NVIDIA GPUDirect RDMA technology, allowing CPU-driven copies between host and GPU… | 83 | 1416 | active |
| RahulSChand/gpu_poor A web-based calculator that estimates GPU memory requirements and inference/finetuning throughput (token/s) for any LLM. It supports quanti… | 19 | 1401 | active |
| TheRock TheRock is AMD's open-source build and release system for the ROCm software stack, replacing the legacy monolithic ROCm release process wit… | 89 | 1394 | active |
| CUDA Core Compute Libraries Thrust is a C++ parallel algorithms library providing an STL-like high-level interface for GPU and multicore CPU computing, built on CUDA, … | 10 | 5001 | maintenance |
| Kratos Multiphysics Kratos Multiphysics is a C++ framework with an extensive Python interface for building parallel, multi-disciplinary simulation software suc… | 91 | 1385 | active |
| bytedance/flux Flux is a GPU kernel library from ByteDance that overlaps computation with communication for tensor and expert parallelism in dense and MoE… | 30 | 1369 | active |
| k2-fsa/k2 k2 is a C++/CUDA library with Python bindings that implements differentiable Finite State Automaton (FSA) and Finite State Transducer (FST)… | 61 | 1368 | active |
| NVIDIA-AI-IOT/torch2trt torch2trt is a Python library that converts PyTorch models to TensorRT engines using the TensorRT Python API, with a simple single-function… | 23 | 4885 | maintenance |
| jfalcou/eve EVE is a C++20 header-only library providing a type-based, high-abstraction wrapper over SIMD instruction sets (SSE/AVX-512, NEON/SVE, Alti… | 68 | 1364 | active |
| alibaba/rtp-llm RTP-LLM is Alibaba's high-performance LLM inference engine written in C++/CUDA, optimized with kernels like PagedAttention and FlashAttenti… | 65 | 1364 | active |
| MegEngine/MegEngine MegEngine is a fast, scalable deep learning framework with automatic differentiation, developed in C++ with Python bindings. It unifies tra… | 23 | 4810 | maintenance |
| LuxCoreRender/LuxCore LuxCoreRender is a physically based, unbiased rendering engine with a C++ and Python API (LuxCore), supporting CPU, OpenCL, CUDA, and OptiX… | 99 | 1339 | active |
| BytedTsinghua-SIA/CUDA-Agent CUDA-Agent is a large-scale agentic reinforcement learning system from ByteDance Seed and Tsinghua that trains LLMs to generate high-perfor… | 53 | 1338 | active |
| facebookincubator/AITemplate AITemplate is a Python framework that compiles deep neural networks into high-performance CUDA (NVIDIA) or HIP (AMD) C++ code for fast fp16… | 67 | 4730 | maintenance |
| nvidia-isaac/nvblox nvblox is a GPU-accelerated C++/Python library for real-time 3D reconstruction using TSDF and ESDF volumetric mapping, designed for robots … | 79 | 1307 | active |
| turboderp-org/exllamav2 ExLlamaV2 is a fast Python inference library for running large language models locally on modern consumer GPUs, with support for EXL2/GPTQ … | 58 | 4631 | maintenance |
| vipshop/cache-dit Cache-DiT is a PyTorch-native inference engine that accelerates Diffusion Transformer (DiT) models with hybrid caching, parallelism, quanti… | 79 | 1287 | active |
| stotko/stdgpu stdgpu is a lightweight C++17 library providing STL-like generic data structures (vector, unordered_map, unordered_set, deque, queue, stack… | 65 | 1270 | active |
| thu-ml/SpargeAttn SpargeAttn is the official implementation of a training-free sparse and quantized attention method that accelerates inference for language,… | 48 | 1257 | active |
| MoonshotAI/FlashKDA FlashKDA is a set of high-performance CUDA kernels (built on CUTLASS) implementing Kimi Delta Attention, a linear attention mechanism, for … | 56 | 1256 | active |
| dmlc/dlpack DLPack is a minimal, stable open in-memory tensor structure defined as a single C ABI header, enabling zero-copy tensor exchange between de… | 73 | 1253 | stable |
| chelsea0x3b/cudarc cudarc is a safe, minimal Rust wrapper around the NVIDIA CUDA toolkit, exposing the CUDA driver API plus libraries such as NVRTC, cuBLAS/cu… | 97 | 1231 | active |
| microsoft/MInference MInference is a Microsoft library that accelerates long-context LLM inference using dynamic sparse attention, reducing pre-fill latency by … | 58 | 1229 | active |
| CNugteren/CLBlast CLBlast is a lightweight, tunable OpenCL BLAS library written in C++11 that implements basic linear algebra subprograms for vectors and mat… | 77 | 1193 | stable |
| getkeops/keops KeOps (pykeops) is a Python library for computing kernel reductions over large arrays on CPUs and GPUs using efficient C++/CUDA routines wi… | 61 | 1192 | active |
| Tencent/hpc-ops HPC-Ops is a production-grade C++/CUDA operator library for high-performance LLM inference, developed by Tencent's Hunyuan AI Infra team. I… | 58 | 1170 | active |
| Dao-AILab/quack QuACK is a collection of high-performance GPU kernels (RMSNorm, LayerNorm, softmax, cross-entropy, GEMM with epilogues) written in NVIDIA's… | 83 | 1167 | active |
| extropic-ai/thrml THRML is a JAX library for building and sampling probabilistic graphical models, focused on efficient block Gibbs sampling of energy-based … | 70 | 1167 | active |
| NVIDIA/cuda-quantum NVIDIA's CUDA-Q platform providing C++ and Python tools for programming heterogeneous quantum-classical workflows across QPUs, GPUs, and CP… | 93 | 1164 | active |
| IST-DASLab/marlin Marlin is a highly optimized FP16xINT4 matrix multiplication CUDA kernel for LLM inference that achieves near-ideal 4x speedups at batch si… | 26 | 1164 | active |
| Linaom1214/TensorRT-For-YOLO-Series A Python and C++ toolkit for running YOLO-series object detection models (YOLOv3 through YOLOv12, YOLOX) with NVIDIA TensorRT, including ON… | 40 | 1158 | active |
| inducer/pyopencl PyOpenCL is a Python wrapper providing Pythonic access to the OpenCL parallel computation API, letting you run kernels on GPUs and other ma… | 97 | 1156 | stable |
| NVIDIA/libnvidia-container A C library and CLI utility (nvidia-container-cli) that automatically configures GNU/Linux containers to leverage NVIDIA hardware, relying … | 90 | 1132 | stable |
| gunrock/gunrock Gunrock is a CUDA/C++ library for GPU-accelerated graph analytics built on a high-level, data-centric abstraction over vertex and edge fron… | 57 | 1102 | active |
| bytedance/byteps BytePS is a high-performance parameter server framework for distributed deep neural network training, supporting TensorFlow, Keras, PyTorch… | 10 | 3717 | maintenance |
| pocl/pocl PoCL (Portable Computing Language) is an MIT-licensed, conformant open-source implementation of the OpenCL 3.0 standard that can be easily … | 85 | 1083 | active |
| NVIDIA-Merlin/HugeCTR HugeCTR is a GPU-accelerated deep learning framework from NVIDIA designed for training and inference of large recommender models, especiall… | 71 | 1073 | active |
| cuOpt NVIDIA cuOpt is a GPU-accelerated optimization engine for linear programming, quadratic programming, mixed integer programming, and vehicle… | 83 | 1067 | active |
| neka-nat/cupoch Cupoch is a C++/Python library that implements rapid 3D data processing for robotics using CUDA, based on Open3D. It provides GPU-accelerat… | 59 | 1062 | active |
| zhuzilin/ring-flash-attention A Python library implementing RingAttention on top of FlashAttention for distributed long-context transformer training. It provides varlen … | 38 | 1061 | active |
| LuisaGroup/LuisaCompute LuisaCompute is a high-performance cross-platform computing framework for graphics and beyond, featuring a C++-embedded DSL for GPU kernel … | 77 | 1055 | active |
| brichard19/BitCrack BitCrack is a C++ command-line tool for brute-forcing Bitcoin private keys using CUDA (NVIDIA) or OpenCL (AMD/Intel) GPU acceleration. It w… | 23 | 1048 | active |
| NVIDIA/raft NVIDIA RAFT (Reusable Accelerated Functions and Tools) is a C++/Python library of CUDA-accelerated algorithms and primitives for machine le… | 90 | 1043 | active |
| NVIDIA/cuda-tile CUDA Tile IR is an MLIR-based intermediate representation and compiler infrastructure for optimizing CUDA kernels with tile-based computati… | 80 | 1035 | active |
| Accelerate Accelerate is an embedded domain-specific language in Haskell for high-performance, multi-dimensional array computations. Programs written … | 59 | 1020 | active |
| NVIDIA/cudf-spark A plugin library for Apache Spark from NVIDIA that leverages GPUs via the cuDF (CUDA DataFrame) library to accelerate Spark SQL and DataFra… | 97 | 1012 | active |
| pytorch/glow Glow is a machine learning compiler and execution engine for neural network hardware accelerators, developed under the PyTorch project. It … | 10 | 3316 | maintenance |
| bytedance/lightseq LightSeq is a high-performance CUDA-based library for training and inference of sequence models like Transformer, BERT, GPT, and BART, with… | 10 | 3292 | maintenance |
| NebuTech/NBMiner NBMiner is a closed-source GPU cryptocurrency miner supporting algorithms like ethash, etchash, kawpow, cuckoo_ae, beamv3, octopus, and erg… | 23 | 3177 | maintenance |
| qwopqwop200/GPTQ-for-LLaMa A Python library that applies GPTQ 4-bit weight quantization to LLaMA large language models, drastically reducing memory usage and checkpoi… | 31 | 3070 | maintenance |
| shouxieai/tensorRT_Pro A C++ library built on NVIDIA TensorRT that provides a high-level interface for high-performance deep learning inference, with Python bindi… | 23 | 2895 | maintenance |
| cybertronai/gradient-checkpointing A Python library that reduces GPU memory usage when training very deep neural networks via gradient checkpointing, trading computation for … | 32 | 2847 | maintenance |
| microsoft/pai OpenPAI is an open-source AI platform from Microsoft that provides resource scheduling and cluster management for machine learning workload… | 68 | 2700 | maintenance |
| turbo/js turbo.js is a small JavaScript library for GPGPU computing in the browser, letting you run parallel kernels written in GLSL over large Floa… | 10 | 2655 | maintenance |
| trexminer/T-Rex T-Rex is a closed-source NVIDIA GPU cryptocurrency miner supporting algorithms like Ethash, Etchash, Autolykos2, Kawpow, and Octopus, with … | 23 | 2640 | maintenance |
| microsoft/DirectML DirectML is a high-performance, hardware-accelerated DirectX 12 library for machine learning that provides GPU acceleration across DirectX … | 55 | 2582 | maintenance |
| GPUOpen-Effects/FidelityFX-FSR AMD FidelityFX Super Resolution (FSR 1.0) is an open-source spatial upscaling solution that produces high-resolution frames from lower-reso… | 23 | 2187 | maintenance |
| BlazingDB/blazingsql BlazingSQL is a GPU-accelerated distributed SQL engine for Python, built on the RAPIDS ecosystem and cuDF GPU DataFrames. It lets users run… | 23 | 2014 | maintenance |
| google/TensorNetwork TensorNetwork is a Python library for building and contracting tensor networks, with backends for TensorFlow, JAX, PyTorch, and NumPy. It p… | 10 | 1871 | maintenance |
| CUDA Core Compute Libraries CUB is a header-only C++ library providing reusable cooperative primitives (sort, scan, reduction, histogram) for device-wide, block-wide, … | 10 | 1843 | maintenance |
| OpenPPL/ppq PPQ (PPL Quantization Tool) is an extensible, high-performance offline neural network quantization framework written in Python. It parses a… | 23 | 1822 | maintenance |
| jkjung-avt/tensorrt_demos A collection of Python demo programs showing how to optimize and run deep learning models (YOLOv4, YOLOv3, SSD, MTCNN, GoogLeNet, MODNet) w… | 46 | 1793 | maintenance |
| facebookresearch/TensorComprehensions Tensor Comprehensions is a C++ library with a Python API that provides a domain-specific language for expressing machine learning workloads… | 10 | 1769 | maintenance |
| GPGPU-Sim GPGPU-Sim is a cycle-level simulator that models contemporary NVIDIA GPUs running CUDA and OpenCL workloads, including support for TensorCo… | 78 | 1750 | maintenance |
| boostorg/compute Boost.Compute is a C++ GPU/parallel-computing library built on OpenCL, providing a thin C++ wrapper over the OpenCL API plus an STL-like in… | 72 | 1663 | maintenance |