Other time constrained scaling examples include search (humans are not willing to wait a long time for results even if better results are available) and robotics (real time or faster then realtime computations allow robots to adapt to the changing world around them).
yes
Another time constrained scaling example might be weather forecasting/simulation overnight, where the results must be available for the next day. Seems like it is less urgent and has a longer time period, but the more accurate the prediction the better.
tarabyte
I am really interested in music. Would being able to analyze sound, compute fft's, and determine pitch so quickly that it seems "real time" be complex enough to benefit from the kind of parallelism we do in this class?
bjay
This raises an interesting question for me, and it's something that we've mentioned in earlier lectures. With GPU's, it seems that we are chiefly concerned with every possible resource being used, as opposed to simply how quickly our program completes (in the case of a GPU, for example, how quickly we can render our scene). I'm interested to see how resource utilization and execution time are correlated
coffee7
@bjay utilization is important because being able to use more resources leads to possibly being able to increase throughput by doing more operations at one time. As we have talked about in class, it is hard to improve execution time by decreasing latency because ILP and processor speeds aren't becoming better at this point in time. Therefore, we need to focus on increasing throughput to get more things done at the same time in parallel.
Other time constrained scaling examples include search (humans are not willing to wait a long time for results even if better results are available) and robotics (real time or faster then realtime computations allow robots to adapt to the changing world around them).
Another time constrained scaling example might be weather forecasting/simulation overnight, where the results must be available for the next day. Seems like it is less urgent and has a longer time period, but the more accurate the prediction the better.
I am really interested in music. Would being able to analyze sound, compute fft's, and determine pitch so quickly that it seems "real time" be complex enough to benefit from the kind of parallelism we do in this class?
This raises an interesting question for me, and it's something that we've mentioned in earlier lectures. With GPU's, it seems that we are chiefly concerned with every possible resource being used, as opposed to simply how quickly our program completes (in the case of a GPU, for example, how quickly we can render our scene). I'm interested to see how resource utilization and execution time are correlated
@bjay utilization is important because being able to use more resources leads to possibly being able to increase throughput by doing more operations at one time. As we have talked about in class, it is hard to improve execution time by decreasing latency because ILP and processor speeds aren't becoming better at this point in time. Therefore, we need to focus on increasing throughput to get more things done at the same time in parallel.