Modern Videoadapter Architectures. GPGPU Technology (Part 2)

• Date: 2013
• Main Authors: Pogorilyy, S. D., Vitel, D. Yu., Vereshchynsky, O. A.
• Format: Article
• Language: Ukrainian
• Published: Інститут проблем реєстрації інформації НАН України 2013
• Subjects: General-purpose graphics processing units (GPGPU); NVidia Compute Unified Device Architecture (CUDA); DirectCompute; OpenCL; Single Instruction Multiple Data (SIMD); Single In¬struction Multiple Threads (SIMT); shaders; Message Passing Interface (MPI); hierarchical thread organi¬zation; synchronization model; memory model; heterogeneous distributed systems; dynamical parallelism; dynamical memory allocation; videoadapter architectures (Tesla; Fermi; Kepler)
• Online Access: http://drsp.ipri.kiev.ua/article/view/103367
• Journal Title: Data Recording, Storage & Processing

General principles of a shared and distributed memory are described in detail for NVidia CUDA technology. Basic patterns of the thread interaction and globals synchronization problems are analyzed. A comparative analysis of the main GPGPU technologies — Nvidia CUDA, OpenCL, Direct Compute — has been provided. Tabl.: 3. Fig.: 3. Refs: 9 titles.