問題文
A team must train a mid-sized image model and then run the trained model on a handheld scanner. The budget allows one machine for training. Why does the training step gain from a graphics unit rather than from a central unit?
選択肢
- The graphics unit holds thousands of lengthy pipelines, so a branch decision spreads over them, and the central unit holds only a few pipelines built for looped work.
- The graphics unit holds thousands of big caches, so a lookup request spreads over them, and the central unit holds only a few caches built for shared use.
- The graphics unit holds thousands of big buffers, so a memory transference spreads over them, and the central unit holds only a few buffers built for cached input.
- The graphics unit holds thousands of tiny cores, so a matrix product spreads over them, and the central unit holds only a few cores built for arithmetic work.