Bam-zine

Number Crunching

December 31, 2008 on 12:05 pm | In Geek Stuff | Comments Off

I have been running Volunteer Computing projects like Seti@home for some time now searching the Cosmos for E.T. With my new computer I can run several applications at once, one on each core of the cpu, and can get much more work done in the same amount of time. Earlier in December the boffins at BOINC updated their program so it can make use of Nvidias CUDA technology. This allows programers to use the GPU cores on your graphics card as maths co-processors. Basically it means you can run even more number crunching work in an even shorter time. Graphics card chips run much faster than your standard CPU, and also have many more processing cores than your normal CPU. If you want to take this to the extreme, you can buy a dedicated box that just contains super fast graaphics chips to run as a plug’n'play desktop supercomputer. I wish I had the cash to splash out on a few of those…

This type of computing isn’t just limited to the search for little green men, but is being used for all sorts of practical research, from climate prediction to study of diseases you can even help with data from the Large Hadron Collider once its up and running again, who knows you could be the first person to discover Higson’s boson particles and go down in history. There is a huge list of research projects that are using BOINC and this form of mass volunteer computing. So head over to BOINC and download the client application and sign up for a few research projects. The program just sits in your Windows Taskbar (also available fore other OS) and can be set to only run when you are not actually using your computer, so it won’t slow your PC down at all.

As this comes under Geek Stuff there has to be more to it than this, and there is. You can mess about with special optimized applications for your specific hardware to maximise the efficiency of your computation. As I said earlier the CUDA application runs work units much faster than normal, up to 10x faster. Using an application compiled specifically for your CPU can also help speed things up. The vanilla Astropulse application used to take around 100 hours of cpu time to process, but since updating my Astropulse application to one that utilises SSE3, SSE4 instructions etc my times for an Astropulse work unit have dropped to about 24 hours!