GCP Generative AI Leader Practice Question 34: Custom TPU Training Advantage
Google Cloud Generative AI Leader · Test 4 · question 4 of 10
An infrastructure team wants to know what gives Google Cloud a distinctive hardware advantage for training and serving large AI models at scale, beyond the GPUs it also offers. What should they point to?
Show answer and explanation
Answer:
TPUs are Google's own custom-developed, application-specific integrated circuits, purpose-built from the ground up to accelerate machine learning and AI workloads. They are a distinctive part of Google Cloud's AI-optimized infrastructure, alongside GPUs, and are used to train and serve large models such as Gemini at scale.
Options A, C, and D name real Google Cloud services, but none of them is AI-specific hardware. Cloud CDN speeds up content delivery, Cloud Load Balancing spreads network traffic across instances, and Persistent Disk provides general-purpose block storage. None of them accelerates the matrix computations at the core of training or serving a large model.
Reference: About TPUs on Google Cloud