AI Infrastructure

⚠GotchaMostly beyond the ACE exam

The exam touches this section at exactly two bullets: identifying whether to use GPUs or TPUs (§2.1) and attaching GPUs and TPUs (§3.1). The GPUs & TPUs page covers both; the depth here is course material, not exam material.

How Google Cloud runs AI/ML workloads: the accelerators (CPU, GPU, TPU), the machine shapes that carry them, and the provisioning, frameworks, and optimizations that keep them busy.

Start with Cloud GPUs - the accelerated-compute story and the GPU options, clusters, frameworks, and cost tuning that sit on top of it.

Then Cloud TPUs - Google's custom ML ASICs: what they are, when to use them, and why to choose them.

Finally AI Hypercomputer - creating and deploying AI clusters on the integrated supercomputing stack, starting with the six-step cluster creation process.