Project Information

Project Name: CUDA Computing

This project is an appraisal of a computing environment based on Nvidia ARM and Intel for COTS and research applications. We will benchmark COTS (commercial off the shelf) applications including a few that have been optimized and certified for Nvidia Maximus (Quadro and Tesla CUDA). We will benchmark scientific and research applications on ARM, Quadro, and CUDA and attempt to fine tune the conversion of applications from running under x86 to ARM computing environment.

Furthermore, Semi-global matching (SGM) is one of the best ways for doing stereo matching in computer vision currently. We would like to have our own experience of applying CUDA to speed up SGM, and the comparison could also be drawn with belief propagation stereo on CUDA.

First of all, we need to know what is a CUDA Computing?

Compute Unified Device Architecture(CUDA) is a parallel computing platform and programming model invented by NVIDIA. It enables dramatic increases in computing performance by harnessing the power of the graphics processing unit (GPU).

an image

The diagram shows how an application that normally runs in the CPU of a PC is ported over to the GPU.

A software program is made up of application codes.  The 1st step is to partition some sections of the code of a repeating nature to run on the GPU and other sections to remain on the CPU. Use C, C++ or other supported languages with special keywords.
The sections of the code allocated to the GPU must be highly computing intensive and they tend to be the core algorithm and thus the critical parts of the application.  A minor effort of code porting will result in significant performance gains.

At the center of this parallel computing model is CUDA (Compute United Device Architecture) which is NVIDIA’s parallel computing hardware and programming model. CUDA provides developer tools that have made parallel programming easy and accessible.

Copyright © 2012. All Rights Reserved.

Btech451 created by HaoYu,Gan.