Exam AWS Certified Machine Learning - Specialty topic 1 question 304 discussion

https://aws.amazon.com/it/blogs/machine-learning/improve-price-performance-of-your-model-training-using-amazon-sagemaker-heterogeneous-clusters/

The topic is wierd. GPU is idle, which means CPU is not able to feed data to GPU in time, which means more CPU is needed. The current ratio is 12:1, and you need to increase the ratio. D is wrong because 6:1 is smaller than 12:1

A: No - removing GPU could significantly increase training time B: No - doesn't solve the issue of GPU under utilization C: No - doesn't solve the issue of GPU under utilization and may take longer D: Yes - Reducing CPU's should solve the issue of GPU underutilization withour causing training delays

ChatGPT D. Switch to an instance type that has a CPU:GPU ratio of 6:1. This solution aligns with reducing training costs without extending the duration of training jobs. By selecting an instance with a lower CPU:GPU ratio, the specialist can ensure more consistent utilization of the GPU, thereby reducing idle time and optimizing resource use without compromising training efficiency.

Switching to an instance type that has a CPU: GPU ratio of 6:1 will reduce the training costs by using fewer CPUs and GPUs, while maintaining the same level of performance. The GPU idle time indicates that the CPU is not able to feed the GPU with enough data, so reducing the CPU: GPU ratio will IT Certification Guaranteed, The Easy Way! 167 balance the workload and improve the GPU utilization. A lower CPU: GPU ratio also means less overhead for inter-process communication and synchronization between the CPU and GPU processes. References: Optimizing GPU utilization for AI/ML workloads on Amazon EC2 Analyze CPU vs. GPU Performance for AWS Machine Learning