NVIDIA/cuda-samples
Cuda Samples
This repository is a collection of reference implementations and programming examples for the CUDA Toolkit. It serves as a GPGPU implementation guide and a parallel computing reference, providing code for using graphics hardware to perform general-purpose calculations and high-performance parallel processing.
The project provides specific samples for GPU kernel development and resource management. These include demonstrations of multi-GPU communication, peer-to-peer memory access, and system hardware inspection to coordinate distributed GPU resources.
The codebase covers a wide range of capabilities, including GPU memory management, performance optimization through execution graphs, and the integration of domain-specific libraries for linear algebra and image processing. It also demonstrates interoperability between compute contexts and graphics APIs to combine rendering and processing tasks.
Features
- GPU Kernel Offloading - Demonstrates the compilation and launch of specialized C-style functions from the host CPU onto the GPU device.
- GPU Kernel Programming - Serves as a primary reference for writing and executing parallel compute kernels on GPU hardware.
- CUDA-Accelerated Libraries - Demonstrates the integration of specialized CUDA libraries for linear algebra, Fourier transforms, and image processing.
- Cooperative Thread Groups - Provides implementations for coordinating sets of threads to synchronize memory access and manage collective workloads.
- CUDA Programming Examples - Provides a comprehensive collection of reference implementations for GPU-accelerated computing using the CUDA Toolkit.
- GPGPU Implementation Guides - Acts as a practical guide for implementing general-purpose calculations on graphics hardware via CUDA.
- GPU Peer-to-Peer Memory Access - Provides implementations allowing multiple GPUs to read and write to each others memory without routing data through the CPU.
- Multi-GPU Resource Coordination - Demonstrates how to coordinate multi-GPU communication and distribute workloads across multiple devices.
- Direct Memory Access - Provides examples of moving data between system RAM and GPU VRAM using high-speed buses to minimize processing latency.
- GPU Memory Allocators - Provides utilities for allocating and moving data between CPU and GPU buffers using direct memory access.
- SIMT Execution Models - Implements the SIMT model to run the same instruction across multiple threads for parallel processing of large datasets.
- GPU Kernel Launch Configurations - Provides examples for executing source code on the graphics processor using high-level interfaces and custom launch configurations.
- GPU Parallelism Strategies - Implements advanced parallelism using cooperative groups and execution graphs to optimize GPU workload distribution.
- GPU-Accelerated Computation - Provides reference implementations for writing and executing parallel programs that accelerate complex calculations on GPU hardware.
- High-Performance and Parallel Computing - Offers a reference for implementing high-performance parallel processing, kernel execution, and thread coordination.
- Unified Memory Systems - Creates a single virtual memory space shared between the CPU and GPU to simplify data movement.
- Graphics Resource Sharing - Provides examples of sharing memory and texture data across compute contexts and graphics APIs.
- GPU Resource Management - Provides samples for managing the lifecycle and allocation of GPU resources and hardware inspection.
- GPU & Performance - Implements techniques to measure memory bandwidth and apply execution strategies to increase graphics hardware processing speed.
- Performance Optimization Samples - Includes demonstrations for measuring memory bandwidth and using execution graphs to optimize performance.
- CUDA Domain Library Examples - Provides reference code for applying domain-specific libraries for linear algebra and image processing.
- GPU Linear Algebra Libraries - Utilizes specialized GPU-optimized libraries for linear algebra and image processing to solve complex mathematical problems.
- Static Task Graphs - Implements sequences of GPU tasks as static graphs to reduce the overhead of repeated kernel launches.
- Graph Execution Compilers - Compiles and executes captured operation graphs as executable objects to optimize GPU task scheduling.
- Hardware Inspection - Provides tools to retrieve technical specifications about available hardware devices and the current execution environment.
- Development Utilities - Sample code for testing CUDA integration within the subsystem.
