In case it's hard to read, green is transistors (DDD), dark blue is clock speed (MHz), light blue is power (Watts), and purple is performance/clock (ILP) rate. ILP is instruction level parallelism, and I believe DDD is Double Diffused Drain...?
This comment was marked helpful 0 times.
choutiy
The answer to the question in this slide is:
parallelism performance is more focused on latency than throughput. Although the dumb way of asking a 10-core processor to do the same at once will increase the throughput by a factor of 10 but not affect the latency, the ultimate goal of parallelism is to shorten the amount of time to finish a certain job, through better algorithms, work distribution and communications.
This comment was marked helpful 0 times.
kayvonf
@choutiy: I'm not sure if I agree entirely. The four-core CPU in my laptop allows me to watch a video on YouTube, run MS Word, download from Netflix, and play a song in iTunes easily in parallel, but it doesn't reduce the time I get any of those things done in comparison to if I was running any one of those programs on its own on a single core processor.
While its true that a single task with ample sub-task parallelism may complete faster on a parallel machine, parallel execution of many tasks at once can actually slow down the time to task any one of those tasks.
In case it's hard to read, green is transistors (DDD), dark blue is clock speed (MHz), light blue is power (Watts), and purple is performance/clock (ILP) rate. ILP is instruction level parallelism, and I believe DDD is Double Diffused Drain...?
This comment was marked helpful 0 times.
The answer to the question in this slide is:
parallelism performance is more focused on latency than throughput. Although the dumb way of asking a 10-core processor to do the same at once will increase the throughput by a factor of 10 but not affect the latency, the ultimate goal of parallelism is to shorten the amount of time to finish a certain job, through better algorithms, work distribution and communications.
This comment was marked helpful 0 times.
@choutiy: I'm not sure if I agree entirely. The four-core CPU in my laptop allows me to watch a video on YouTube, run MS Word, download from Netflix, and play a song in iTunes easily in parallel, but it doesn't reduce the time I get any of those things done in comparison to if I was running any one of those programs on its own on a single core processor.
While its true that a single task with ample sub-task parallelism may complete faster on a parallel machine, parallel execution of many tasks at once can actually slow down the time to task any one of those tasks.
This comment was marked helpful 0 times.