If the ratio of math instructions to load instructions increases, the computations will be able to "hide" loading latency and increase throughput. This is because more workers can do more useful computation work at any point in time while workers that are loading data can be swapped off the processor until they finish, thereby resulting in shorter/less stalls.
If the ratio of math instructions to load instructions increases, the computations will be able to "hide" loading latency and increase throughput. This is because more workers can do more useful computation work at any point in time while workers that are loading data can be swapped off the processor until they finish, thereby resulting in shorter/less stalls.