A data scientist considers a GPU for a large neural-network training workload. Which advantage is most relevant?

1. Automatic removal of all incorrect labels
2. Elimination of the need for a training objective
3. A guarantee of perfect model generalization
4. Efficient parallel execution of many numerical operations