MLS-C01 Question 165
Single answerYou are training a deep learning model using a large image dataset with millions of high-resolution images. The model architecture is a convolutional neural network (CNN) with multiple layers and requires extensive matrix computations. Training speed is a priority to meet a tight project deadline. Which compute resource is the most appropriate for this scenario?
- A
A CPU instance with high memory (e.g., r5.xlarge)
- B
A GPU instance optimized for machine learning (e.g., p4d.24xlarge)
- C
A compute-optimized CPU instance (e.g., c5.4xlarge)
- D
An instance with Elastic Inference enabled to reduce costs
Show answer and explanation
Correct answer: B
Explanation
Training deep learning models, especially convolutional neural networks (CNNs), involves extensive matrix and tensor computations that are highly parallelizable. GPU instances optimized for machine learning, like the p4d.24xlarge, are specifically designed to handle these workloads efficiently, significantly reducing training time compared to CPU instances. This makes GPUs the most appropriate choice in scenarios where training speed is critical and the dataset is large.
- A. Incorrect.
While CPU instances with high memory can handle large datasets, they are not optimized for the parallel computations required for training deep learning models, especially convolutional neural networks (CNNs). Training on CPUs would be significantly slower compared to GPUs.
- B. Correct.
GPU instances optimized for machine learning, such as p4d.24xlarge, are specifically designed for tasks that require extensive parallel computations, such as training deep learning models. GPUs can significantly speed up training times for CNNs with large datasets.
- C. Incorrect.
Compute-optimized CPU instances, such as c5.4xlarge, are suited for high-performance computing tasks but lack the parallel processing capabilities of GPUs. They are not the best option for training deep learning models efficiently.
- D. Incorrect.
Elastic Inference is designed to reduce costs during inference by attaching GPU acceleration to CPU instances. It is not applicable for speeding up the training of deep learning models and is therefore not suitable for this scenario.