GPGPU-вычисления и архитектура CUDA

Уровень 0 · материалов: 4

В кластер входят документы, посвященные техническим принципам, архитектуре и реализации вычислений общего назначения на графических процессорах.

Общие признаки: параллельное программирование на GPU, архитектура NVIDIA CUDA, переход от CPU к GPU, ускорение вычислений общего назначения

Группа выше: Вычисления на GPU и графические стандарты

Материалы

• #22426 — GPU Computing: Why, When, and How to Use It — Представитель этого кластера

  Вычисления на GPU – зачем, когда и как. Плюс немного тестов — DenisT

  Смысл: The main idea is to demystify GPU computing for non-game developers by explaining the architectural shift to SIMD, the trade-off between compute power and data transfer overhead, and the practical tools available to implement GPGPU acceleration.

  A technical guide explaining when and how to offload computations to a GPU, highlighting the necessity of SIMD-compatible algorithms and the impact of data transfer overhead.

• #10961 — Understanding GPU Computing with NVIDIA CUDA

  CUDA: Как работает GPU — MaxFX

  Смысл: The text aims to explain the fundamental software and hardware architecture of NVIDIA GPUs using the CUDA platform, demonstrating how to leverage parallel processing for mathematical computations.

  A detailed technical guide explaining NVIDIA CUDA's computational model, API structure, and a practical implementation of parallel vector addition on a GPU.

• #33 — Introduction to Parallel Programming with CUDA

  Параллельное программирование с CUDA. Часть 1: Введение — VladGorbatiuk

  Смысл: The main idea is to introduce the basic concepts of CUDA parallel programming, emphasizing the transition from latency-oriented CPU computing to throughput-oriented GPU computing through practical examples of image processing.

  An introductory guide to CUDA that explains the host-device model and demonstrates how to optimize GPU kernels for image processing to achieve higher throughput than CPUs.

• #24974 — Developing a GPU-Based Physics Game

  Доделал игру, работающую на видеокарте — ThisIsZolden

  Смысл: The text shares the technical journey of creating a game based on GPGPU (General-Purpose computing on Graphics Processing Units), weighing the massive performance gains of parallel processing against the complexities of hardware compatibility, API limitations, and the steep learning curve.

  A developer details the pros and cons of building a physics-heavy game using Unity compute shaders, highlighting extreme performance gains versus significant cross-platform and hardware challenges.