In the previous slide we demonstrated the benefit of splitting one powerful core into two slightly less powerful ones. In theory, since we have two cores running at 75% of the speed of the single core we should have 1.5x speedup on the tasks that we do. However, in this example, the code runs as one thread and can only be run on one core at a time. Therefore, if we were to run this on our "faster" two core setup, it will run 25% slower.
This relates directly to the first lecture where we talked about how software has to be parallel to see performance gains from multiple threads and cores.
This comment was marked helpful 0 times.
snkulkar
Even though we have 2 cores running at 0.75 times the speed of the original core, when we run this program we won't get a 1.5 x speedup. This is because the compiler is not capable of recognizing that everything inside the main for loop are independent instructions and could be done in parallel.
This slide makes the point that just adding more cores does not necessarily make a given program faster. The programmer must understand which instructions are capable of being executed simultaneously and thus, exploit the multiple cores.
In the previous slide we demonstrated the benefit of splitting one powerful core into two slightly less powerful ones. In theory, since we have two cores running at 75% of the speed of the single core we should have 1.5x speedup on the tasks that we do. However, in this example, the code runs as one thread and can only be run on one core at a time. Therefore, if we were to run this on our "faster" two core setup, it will run 25% slower.
This relates directly to the first lecture where we talked about how software has to be parallel to see performance gains from multiple threads and cores.
This comment was marked helpful 0 times.
Even though we have 2 cores running at 0.75 times the speed of the original core, when we run this program we won't get a 1.5 x speedup. This is because the compiler is not capable of recognizing that everything inside the main for loop are independent instructions and could be done in parallel.
This slide makes the point that just adding more cores does not necessarily make a given program faster. The programmer must understand which instructions are capable of being executed simultaneously and thus, exploit the multiple cores.
This comment was marked helpful 0 times.