Getting the 100x speedup from S=0.01 is hard; even with 1000 processors the speedup is only ~90x
This comment was marked helpful 0 times.
acappiello
There are some interesting implications made apparent by this graph.
As we discussed at the very beginning of the course, parallelism was born out of our inability to continue to push the clock speed of processors any higher due to issues with heat (lecture 1, slide 27). However, this demonstrates that there is an upper limit to parallelism, as programs are not 100% parallel. So, while more processors continues to increase the amount of parallelism possible on a system, it becomes harder to exploit it.
Later in the course, we learn that it is important to take into account the capability of each core, and in heterogenous systems, the type of each core.
Getting the 100x speedup from S=0.01 is hard; even with 1000 processors the speedup is only ~90x
This comment was marked helpful 0 times.
There are some interesting implications made apparent by this graph.
As we discussed at the very beginning of the course, parallelism was born out of our inability to continue to push the clock speed of processors any higher due to issues with heat (lecture 1, slide 27). However, this demonstrates that there is an upper limit to parallelism, as programs are not 100% parallel. So, while more processors continues to increase the amount of parallelism possible on a system, it becomes harder to exploit it.
This comment was marked helpful 0 times.
Here is a nice demonstration of Amdahls Law.
This comment was marked helpful 1 times.
Later in the course, we learn that it is important to take into account the capability of each core, and in heterogenous systems, the type of each core.
This comment was marked helpful 0 times.