Introduction to Parallel Programming: Predicting Performance Share your comment!

The video lecture offers definitions for the performance metrics speedup and efficiency. A fence painting example illustrates how to compute these metrics. You’ll also see explanations of why Amdahl’s Law is overly optimistic in the prediction of possible speedups. 

Posted on by Dr. Clay Breshears, Intel®
2 comments
colinmaharaj
colinmaharaj like.author.displayName 1 Like

Very nice. I like the part of added overhead. I was going to use cores in another PC i.e. over a network / cluster, but now I need to add n/wdelays as part of my timing. I guess the amount of time required to calculate my task on a core, has to be much larger than the propagation delays across the n/w or else do it local in hyper-thread.

michellesu
michellesu

I don't  follow Predicting Performance Introduction about  the definitions for the performance metrics speedup and efficiency and Amdahl’s Law, that the efficiency drawing in add cores and speedup curves drawing in add cores are wrong drawing and introduction.