Built on the foundation of NVIDIA® CUDA® , NVIDIA CUDA-X™ is a powerful suite of libraries designed to deliver industry-leading GPU acceleration across AI and high-performance computing use cases—from generative AI and autonomous machines to climate modeling and financial forecasting. Whether you're deploying on resource-constrained IoT devices or the world's largest supercomputers, NVIDIA CUDA-X libraries provide highly optimized implementations of complex algorithms that far outperform CPU-only alternatives. For developers looking to build or scale applications, CUDA-X offers the most efficient and accessible path to maximizing hardware performance across any domain. CUDA Math Libraries Scientific Computing Libraries Physics Libraries Quantum Computing Libraries Deep Learning Core Libraries Parallel Algorithm Libraries Data Processing Libraries Image and Video Libraries Communication Libraries Partner Libraries CUDA Math Libraries GPU-accelerated math libraries lay the foundation for compute-intensive applications in areas such as molecular dynamics, computational fluid dynamics, computational chemistry, medical imaging, and seismic exploration. GPU-accelerated basic linear algebra (BLAS) library. GPU-accelerated library for Fast Fourier Transform implementations. GPU-accelerated random number generation. GPU-accelerated dense and sparse direct solvers. GPU-accelerated BLAS for sparse matrices. GPU-accelerated tensor linear algebra library. GPU-accelerated direct sparse solver library. GPU-accelerated standard mathematical function APIs. GPU-accelerated linear solvers for simulations and implicit unstructured methods. Enabling GPU-accelerated math operations for the Python ecosystem. nvmath-python (Beta) is an open source library that provides high-performance access to the core mathematical operations in the NVIDIA math libraries. Scientific Computing Libraries For applications requiring neural networks that respect mathematical symmetries—specifically, equivariance with respect to 3D geometric data, such as molecular structures, proteins, and materials. An open source Python library designed to accelerate the construction and execution of geometry-aware neural networks, particularly those handling data rotations and translations in 3D space. A collection of domain-specific NVIDIA NIM™ microservices and toolkit for accelerating chemical and materials discovery (e.g., battery materials, catalysts, OLEDs, beauty formulations). Targeting the modern-day challenges of nanoscale computational lithography, this library optimizes tools and algorithms to accelerate computational lithography and the manufacturing of semiconductors using GPUs. Accelerates industrial-scale quantum chemistry via a flexible API and high-performance building blocks for first-principles electronic structure calculations on GPUs. Physics Libraries GPU-accelerated physics libraries and frameworks to speed up simulations across domains including computational physics, multiphysics, quantum physics, and weather modeling. A purpose-built, open source Python framework that delivers GPU acceleration for computational physics, AI, and optimization workflows, enabling kernel-based programs for simulation AI, robotics, and ML. NVIDIA PhysicsNeMo An open source Python framework for building, training, and fine-tuning AI physics models at scale. A comprehensive family of open models, libraries, and frameworks that democratize global access to professional-grade weather and climate AI. Quantum Computing Libraries Enabling simulation, HPC integration and AI for quantum computing. A set of highly optimized libraries for accelerating quantum computing simulations. SDK of optimized libraries for accelerating post-quantum cryptography (PQC) workflows. Libraries for simulating and implementing noise-resilient quantum algorithms and error mitigation. GPU-accelerated solvers for hybrid quantum-classical optimization and variational workloads. Deep Learning Core Libraries GPU-accelerated libraries for deep learning applications that use CUDA and specialized hardware components of GPUs. GPU-accelerated library of deep neural network building blocks ("primitives") for deep learning. NVIDIA TensorRT™ and TensorRT LLM High-performance deep learning inference optimizer and runtime for production deployment. Modular C++ templates and Python DSLs for building high-performance kernels targeting NVIDIA Tensor Cores. GPU-accelerated kernel library, accessible via Python API for inference, optimizing attention, MoEs, GEMMs, comms, and other neural network operations. Parallel Algorithm Libraries The CUDA Core Compute Libraries (CCCL) provide GPU-accelerated algorithms in C++ and Python. They provide optimized parallel primitives to solve complex challenges in natural sciences, logistics, travel planning, and more.

Powerful data-parallel library based on the C++ STL that lets developers implement complex GPU-accelerated algorithms in a high-level API without sacrificing performance. Cooperative primitives for CUDA kernel authoring, providing warp-wide, block-wide, and device-wide collective primitives across the CUDA programming model. Pythonic interface for high-performance, device-level CCCL algorithms, enabling developers to leverage CUDA parallel processing directly within Python workflows. Standardized primitives for distributed and local parallel patterns such as sort, scan, and reduction, optimized for the latest NVIDIA GPU architectures. Data Processing Libraries GPU-accelerated libraries to accelerate data processing workflows for tabular, text, and image data. Accelerate tabular data, including pandas, Polars, and Apache Spark with zero code changes. Accelerate vector search for data mining and semantic search applications—including world-class performance from the GPU-native nearest neighbors algorithm CAGRA. Speed up ML algorithms in scikit-learn, UMAP, HDBSCAN, and Apache Spark with zero code changes. Open source, GPU-accelerated decision optimization engine designed to tackle large-scale problems with millions of variables and constraints, enabling accelerated decision-making. Scale up and speed up graph analytics with GPU-accelerated NetworkX. Improves generative AI model accuracy by processing text, image, and video data at scale for training and customization, with pre-built pipelines for generating synthetic data. Open application framework that optimizes cybersecurity AI pipelines for analyzing large volumes of real-time data. High-throughput GPU-accelerated compression and decompression library that minimizes storage footprint and speeds up data transfer rates for AI training, HPC, data science, and analytics applications. GPU Direct Storage NVIDIA GPUDirect Storage creates a direct data path between local or remote storage, such as NVMe or NVMe over Fabrics (NVMe-oF), and GPU memory. Expand data science pipelines to multiple nodes with NVIDIA RAPIDS on Dask. Image and Video Libraries GPU-accelerated libraries for image and video decoding, encoding, and processing that use CUDA and specialized hardware components of GPUs. GPU-accelerated image codec library with a unified interface for high-throughput image encoding and decoding, built as an extensible framework for a wide array of codec plugins. GPU-accelerated library for data loading and pre-processing to accelerate deep learning applications for image, video, and audio modalities. Open source library for high-performance, GPU-accelerated pre- and post-processing in vision AI pipelines. Open source, accelerated computer vision and image processing library for multidimensional images in biomedical, geospatial, material, healthcare, and remote sensing use cases. NVIDIA Performance Primitives (NPP) GPU-accelerated library of highly optimized primitives for CUDA-based 2D image and signal processing, including filtering, color conversion, and image manipulation. NVIDIA Video Codec SDK Hardware-accelerated video encode and decode on Windows and Linux. NVIDIA Optical Flow SDK Exposes the latest hardware capability of NVIDIA GPUs dedicated to computing the relative motion of pixels between images. Communication Libraries Performance-optimized multi-GPU and multi-node communication primitives. Based on the OpenSHMEM one-sided communication model, provides a partitioned global address space across GPU memories. Open source library for fast multi-GPU, multi-node communication that maximizes bandwidth while maintaining low latency. Low-latency inference transfer library, moving KV cache and tensors between GPUs, memory tiers, and storage. Partner Libraries GPU-accelerated open-source library for computer vision, image processing, and machine learning, now supporting real-time operation. Open-source multimedia framework with a library of plug-ins for audio and video processing. GPU-accelerated open-source library for matrix, signal, and image processing. GPU-accelerated linear algebra routines for heterogeneous architectures, by Magma. IMSL Fortran Numerical Library GPU-accelerated open-source Fortran library with functions for math, signal and image processing, and statistics, by RogueWave. Library for graph-processing designed specifically for the GPU. GPU-accelerated functions for sparse direct solvers, included in the SuiteSparse linear algebra package, authored by Prof. Triton Ocean SDK Real-time visual simulation of oceans, water bodies in games, simulation, and training applications, by Triton. Primitives for accelerating imaging applications in medical, industrial, and defense domains. Open source array library for GPU-accelerated computing with Python, providing a NumPy/SciPy-compatible interface. Members of the NVIDIA Developer Program get early access to all CUDA library releases and the NVIDIA online bug reporting and feature request system. Join the Developer Program