MLS-C01 Question 168
Single answerYou are training a deep learning model for image classification on a dataset containing millions of high-resolution images. The model architecture involves convolutional neural networks with a large number of parameters. The training needs to be completed as quickly as possible to meet a tight deadline. Which compute resource configuration is the most appropriate for this workload?
- A
A single CPU instance with high memory
- B
A distributed CPU cluster with multiple instances
- C
A single GPU instance with high memory
- D
A distributed GPU cluster with multiple GPU instances
Show answer and explanation
Correct answer: D
Explanation
For training deep learning models with large datasets and a tight deadline, GPUs are preferred over CPUs due to their ability to perform parallel processing efficiently. Using a distributed GPU cluster allows the workload to be divided across multiple GPUs, significantly reducing training time and meeting the performance requirements. This makes Option 4 the most appropriate choice for the given scenario.
- A. Incorrect.
A single CPU instance with high memory is not suitable for training deep learning models with large datasets and computationally expensive architectures, as CPUs are slower for such operations compared to GPUs.
- B. Incorrect.
A distributed CPU cluster with multiple instances can handle large datasets, but CPUs lack the parallel processing capabilities needed for deep learning training, making this option less efficient.
- C. Incorrect.
A single GPU instance with high memory is more suitable for deep learning workloads than CPUs due to the GPU's parallel processing capabilities. However, for a large dataset and tight deadline, a single GPU may not provide the required speed.
- D. Correct.
A distributed GPU cluster with multiple GPU instances is the best option for this scenario. Distributed training across multiple GPUs accelerates the training process significantly and is well-suited for large datasets and complex models.