Previous | Next --- Slide 4 of 81
Back to Lecture Thumbnails
dasteere

If there is so much more computational power on a gpu why don't we run all large tasks on GPUs and leave the CPU for smaller tasks?

coffee7

@dasteere I think one consideration is memory latency. An example of this is assignment 2: we compared an implementation of saxpy on the CPU and the GPU, and overall performance was about the same even though the GPU has higher memory bandwidth. The reason for the lack of performance increase when using the GPU was having to copy over all of the data to the device.