During development, a data scientist uses a cloud notebook to run light feature-engineering tasks during business hours. A data engineer uses the same notebook about once per day to run heavy memory-intensive preprocessing (about 2 hours); preprocessing does not use GPUs. Currently both run on a general-purpose ml.m5.4xlarge notebook instance. The monthly AWS budget has been exceeded. Which change will yield the greatest cost savings?

The correct option offers the greatest cost savings by right-sizing the resources for each task. The data scientist's light feature engineering can run on a smaller, general-purpose instance, which is more cost-effective than the current ml.m5.4xlarge. Stopping the notebook when idle further reduces costs by only paying for compute when in use. The heavy, memory-intensive preprocessing is best handled by a managed processing job on a memory-optimized r5 instance. This decouples the compute, allowing the preprocessing job to run on an appropriately sized instance only when needed, without tying up the notebook instance. Incorrect options: Switching to a memory-optimized instance for both tasks is inefficient because the feature engineering doesn't require memory optimization, and the preprocessing would still run on a continuously provisioned notebook instance, which is more expensive than a managed job. Keeping the existing instance and using a GPU-accelerated instance for preprocessing is incorrect because the problem explicitly states preprocessing does not use GPUs, making a GPU instance unnecessary and more expensive. Using a reserved memory-optimized instance for preprocessing is less flexible and potentially more expensive than a managed processing job for infrequent, heavy tasks, as reserved instances are better suited for continuous, predictable workloads.