I’m currently in the middle of optimizing some summery generation application which is at the moment very CPU consuming. Basically, the task of this particular application is to monitor a set of servers and gather data published as events and finally generating a summeries of the data after analyzing them. While I was searching for such optimization techniques I thought CUDA would be somewhat useful for my task. But I’ve got a couple of doubts in my mind whether it is the best fit upon my requirement as it does not make sense to make all the tasks parallel but only a portion of them. I would appreciate if anyone could help me decide on this.
Need WAY more information to help you. On top of that, odds are you are the only one who can help yourself. It’s a balancing act between how much data you have (and therefore how much memory transfer you will incur…but also how much you’ll benefit from parallel execution), and how parallelizable the tasks are.