Students at some of the top universities in China have come up with ways to use CUDA, NVIDIA’s parallel computing architecture, to find improvements in areas like pattern matching, image computation and particle simulation.
As part of NVIDIA’s CUDA University Roadshow, students at universities across China were invited to compete for prizes in CUDA programming. Nearly 800 students took up the challenge, producing more than 70 entries over five months. Four judges – experts on CUDA, who’ve taught it at universities in China or used it in their research – decided on the winners.
Four students from South China University of Technology took home the grand prize of 20,000 renminbi ($2,932) for a project that demonstrated a 20x improvement in the speed of matching web pages using an algorithm transplanted to and optimized for GPUs.
Other winning entries include a toolkit for GPU-accelerated linear image registration (with potential for improving medical image analysis) and a solution for speeding up large sparse matrix-vector multiplication (used in advanced science and engineering computations). Whether they were working on gene interactions or radar imaging technology, the winners found often-dramatic performance increases using GPUs and CUDA as compared to traditional CPUs.
CUDA is gaining more exposure in China, with the first Chinese publication of a CUDA programming guide coming in October. It’s clear Chinese programmers and engineers get CUDA’s advantages. Many of the winning entries have potential business value and could turn into real-world solutions with further development.
All the entries along with documents and source code are on NVIDIA’s China CUDA site, where they’re available in Chinese. Until your Mandarin gets up to speed, you can always check out the latest CUDA developments in NVIDIA’s CUDA Zone.