GPU Computing Overview and GPGPU Concepts

GPU Computing Overview

• GPU stands for Graphics Processing Unit
• GPUs are highly parallel specialized circuits for computer graphics
• Examples of GPU devices include: NVIDIA Geoforce GeForce GTX 680, AMD Radeon HD 6870
• GPUs are used in game consoles, workstations/servers, smartphones, PCs, and embedded systems
• GPUs have seen significant performance increases in GFLOPS (giga floating-point operations per second) since 2002
• Recent years show a growing performance gap between GPUs and CPUs
• The GPU architecture, as opposed to a CPU, is designed for high parallel computations
• GPUs perform complex rendering calculations.
• GPUs contain many cores unlike CPUs which have a smaller number of cores.

What is GPGPU?

• GPGPU stands for General-Purpose computing on a Graphics Processing Unit
• Utilizing GPU hardware for non-graphic computations
• Researchers originally noticed GPUs' raw performance for solving compute-intensive problems in science and engineering.
• GPUs were not originally designed to handle non-graphic computations; their original design matched graphics pipeline features.
• Programmers need to map problems onto GPU architecture in order to make use of the parallel nature of the device, running multiple instances of compute functions.

Graphics Pipeline

• The GPU uses a pipeline to process graphics data, progressing through stages.
• The pipeline includes stages such as Vertex Shader, Geometry Shader, Clipping, Screen Mapping, Triangle Setup, Triangle Traversal, Pixel Shader, and Merger stages.
• Three-dimensional (3D) objects and related data are initially processed then mapped to the screen.

Texture Mapping

• Texture mapping is used to apply textures to 3D models.
• Images are transformed onto the 3D surfaces using image data
• Coordinates map image data to surfaces, based on the texture image.

Anti-aliasing

• Anti-aliasing techniques smooth out jagged edges in images.
• An example is shown of reducing jaggedness.

Brief History of GPU Computing

• Early 80's to late 90's: fixed-function graphics pipelines
• Early 2000's: evolution of programmable real-time graphics
• 2006: unified graphics and computing processors
• Key examples: Fermi, Kepler, Tesla architectures. .

GPU Architecture

• GPUs have a different architecture, in comparison to CPUs, consisting of many processing threads.
• GPUs (such as Tesla) have an architecture consisting of multiple streaming multiprocessors (SMs)
• These multiprocessors are linked in clusters
• These clusters can access main memory.
• Data is processed on the cluster, across multiple cores.

CPU-GPU Boundary

• The CPU and GPU interact at the CPU-GPU boundary.
• The boundary handles command and data streaming to perform tasks.
• Commands, data, and results are exchanged between the CPU and GPU.

Diagram Descriptions

• Diagrams show the architecture of the GPU including memory partitions, processing units, and connection to the host computer.
• Various components within a GPU such as L1/L2/cache, vertex shader, texture cache, and frame buffer.
• The diagrams illustrate the flow of data from input to output, and relationships in the GPU.

Explore the fundamentals of GPU computing and the concept of General-Purpose computing on a Graphics Processing Unit (GPGPU). This quiz covers various applications, architecture differences, and performance trends of GPUs in modern computing. Test your knowledge on how GPUs are utilized beyond graphics processing.