Google Professional Machine Learning Engineer Question 288
Single answerGoogle Cloud PlatformYou are designing a machine learning workflow for real-time video analytics. The solution requires low latency inference and high computational power to process frames from multiple video streams simultaneously. Which compute and accelerator option should you choose?
- A
CPUs with autoscaling enabled in a Google Kubernetes Engine (GKE) cluster
- B
GPUs in a Google Kubernetes Engine (GKE) cluster
- C
Cloud TPUs in a Google Compute Engine instance
- D
Google Cloud Edge TPU devices
Show answer and explanation
Correct answer: D
Explanation
For real-time video analytics with low latency requirements, edge devices like Google Cloud Edge TPUs are the most suitable option. Unlike CPUs, GPUs, or Cloud TPUs, Edge TPUs are purpose-built for on-device inference and can handle high computational loads efficiently without relying on cloud connectivity. This makes them ideal for scenarios requiring near-instantaneous processing at the edge.
- A. Incorrect.
CPUs are not optimal for this use case since they lack the parallel processing power needed for real-time video analytics at scale. While autoscaling can address workload spikes, it does not meet the low latency and high computational requirements.
- B. Incorrect.
GPUs are a better choice than CPUs for video analytics due to their parallel processing capabilities. However, they may not be ideal for real-time scenarios at the edge because they are primarily designed for cloud-based or data center workloads.
- C. Incorrect.
Cloud TPUs are powerful for large-scale training and inference workloads in the cloud but are not optimized for edge scenarios requiring low latency and real-time processing.
- D. Correct.
Google Cloud Edge TPU devices are specifically designed for real-time inference at the edge. They are optimized for low latency, high throughput, and on-device processing, making them the best choice for this use case.