GPU Computing Lecture 1 PDF

GPU Computing LO47 [email protected] 1  Overview A. What is GPU? B. What is GPGPU? 2 A. What is GPU?  What is GPU ?  GPU : Graphics Processing unit - Highly parallel specialized circuit for computer gr...

GPU Computing LO47 [email protected] 1  Overview A. What is GPU? B. What is GPGPU? 2 A. What is GPU?  What is GPU ?  GPU : Graphics Processing unit - Highly parallel specialized circuit for computer graphics. NVIDIA Geoforce GeForce GTX 680 AMD Radeon HD 6870 - Game consoles, WS/Servers, Smartphones, PC, embedded systems, … 4 Borrowed from GPGPU Workshop slides (SIGGRAPH 05) 5 Enlarging performance gap between GPUs and CPUs. 6  What is GPU computing ?  Parallel-core GPU works in tandem with multi- core CPUs : - as a purpose-built device able to assist in performing complex rendering calculations - as a separated dedicated graphics resources including graphics processor (Multiple ALUs) and memory (Fast onboard memory) - as a co-processor to accelerate through processing of large blocks of data and portions of the code 7 Source: http://www.nvidia.com/object/what-is-gpu-computing.html 8 9 The Overall System Architecture of a PC, Source: from GPU Gems 2 (NVIDIA), the architecture of the GeForce 6 SeriesGPUs. 10 Brief history 3D Array of pixels 12  Computer graphics Vocabulary Pixel smallest addressable screen element Vertex the corner of a polygon (usually that polygon is a triangle) Polygon a 2D shape, its position is determined by the coordinates of its vertices and can be colored, shaded and textured scene file with geometry, viewpoint, texture, lighting, and shading information Rendering the process of generating a 2D image from a 3D model Rasterization a rendering algorithm often carried out by fixed function hardware within the graphics pipeline along a rendering device, such as a GPU graphics pipeline The graphic system generates images through a pipelined sequence of operations or stages such as Geometric operations and Raster operations (OpenGL state machine, Direct3D) 13 14 GPU and the Graphics pipeline 15 16  Switch to pdf file 17 Texture mapping example: painting a world map texture image onto a globe object. 18 Example of antialiasing operations. 19  Brief history  From early 80’s to late 90’s, the era of Fixed-Function Graphics Pipelines  From 2001, Evolution of Programmable Real-Time Graphics  From 2006, Unified Graphics and Computing Processors  Fermi  Kepler  Tesla 20  The pipeline Source: http://http.developer.nvidia.com/GPUGems/gpugems_ch28.html 21 22 23 24 B. What is GPGPU?  What is GPGPU ?  General-Purpose computing on a Graphics Processing Unit  Using graphic hardware for non-graphic computations 26  What is GPGPU ?  some researchers (and Sony and NVIDIA) took notice of the raw performance of GPUs to solve compute-intensive science and engineering problems  however, GPUs had been designed to match the features required by the graphics pipeline  To access the computational resources, a programmer has to map the original problem to the GPU architecture, i.e., to run many simultaneous instances of a compute function 27 Source: http://www.mzahran.com, CSCI-GA.3033-012 28  GPU Computing  think of the GPU like a processor.  NVIDIA developed the CUDA C/C++ compiler, libraries, and runtime software  TESLA Architecture 29 30 TESLA Architecture, Source: S.Collange, 2010 , 31  TESLA Architecture Source: S.Collange, 2010 32 The common architectural landscape SMP Clusters SMP Node SMP Node PEs, PEs, PEs, PEs,......... Memory Memory Interconnect Massively Parallel Processor Systems IBM Blue Gene Cray XT5 33