If the new performance means the performance we programmer want to get from parallelism, I think we are focusing on throughput using parallelism since the clock speed can not be easily increased.
ote
@yikesaiting I think you are correct because as mentioned in this slide, increasing clock speed cannot be easily increased due to the consumption of power and resulting heat.
stee
I agree. Latency is the time it takes for, say, a thread to complete a task, which is limited by clock frequency. However, having multiple cores and other parallism, we can work on multiple tasks at the same time, thereby improving throughput.
illuminated
I agree with the above comments that the new focus is on throughput due to the limitations of increasing clock speed. However note that for certain tasks improving throughput can also reduce program latency (overall time to complete a program, not an individual memory operation or small component of the task) due to effective performance optimization.
If the new performance means the performance we programmer want to get from parallelism, I think we are focusing on throughput using parallelism since the clock speed can not be easily increased.
@yikesaiting I think you are correct because as mentioned in this slide, increasing clock speed cannot be easily increased due to the consumption of power and resulting heat.
I agree. Latency is the time it takes for, say, a thread to complete a task, which is limited by clock frequency. However, having multiple cores and other parallism, we can work on multiple tasks at the same time, thereby improving throughput.
I agree with the above comments that the new focus is on throughput due to the limitations of increasing clock speed. However note that for certain tasks improving throughput can also reduce program latency (overall time to complete a program, not an individual memory operation or small component of the task) due to effective performance optimization.