CUDA C Research / Project Ideas

Question

Over the summer, I started to learn CUDA C because the nVIDIA performance claims were simply unbelievable. This past week, I started another semester of my undergrad studies. My major is computer science.

One of the classes I am taking this semester is undergrad research and want to further practice with CUDA C. Does anyone have any ideas on what I can work on? My goal of doing the research is to further learn about CUDA C and then by the end of the research have a better understanding when CUDA C should be used in solving problems.

Thanks.

Answer 1

Take some compute intensive task and try a CUDA implementation?
If you are interested in image processng try something from OpenCV - you have the C ( and SSE assember) for the algorithms, converting into CUDA analysing where you get a performance gain - and how it scales, should be a good project. If you want something more ambitious a CUDA implementation of parts of FFMPEG would be useful.

ps. Remember in an ugrad project you are being marked on the analysis of the algorithm, not the coding standards - so don't get carried away with writing huge amounts of infrastructure/library code.

Answer 2

How about you look at a typical problem discussed in the Numerical Recipes book. Then pick a linear algebra problem such as the LU decomposition and implement it using CUDA. Benchmark it and try to understand your results.

Answer 3

Thanks very much, Martin. Looking at it that way is very beneficial to look at iy this way

Answer 4

Maybe try to implement some kind of optimization or data fitting routines. These are very easy to parallelize, and getting 20x speedups is quite easy.