Skip to main content

AI Hypercomputer cheat sheet

Exam guide§2.1

A one-page review card for AI Hypercomputer deployment types. For the full detail behind each item, follow the links back to Cluster creation process, Creating a cluster with Compute Engine, and Building with GKE.

Four main deployment options​

Control vs overheadDirect management (GCE)Maximum control, high overhead.Foundational (GKE)Balances control and automation; ideal for Kubernetesexperts.Open frameworks via ToolkitsLeverages best practices and simplifies complexsetups (e.g. Cluster Toolkit).Fully managed (Vertex AI)Easiest to use; Google handles infrastructure, lessgranular control.
Four deployment options, from most control (GCE) to most managed (Vertex AI).
GotchaToolkits are a distinct option, not just "GKE"

The at-a-glance list splits Open frameworks via Toolkits out from Foundational (GKE). Cluster Toolkit sits on top of the orchestrated path - it ships example blueprints you can customize, and those blueprints reference supported modules that break down into types like scheduler and storage.

Six steps to create a cluster​

NumbersCluster creation, at a glance
  1. Choose machine type
  2. Choose consumption option
  3. Choose deployment option
  4. Choose orchestrator
  5. Choose image
  6. Create and deploy

See Cluster creation process for the machine-type, consumption, deployment, orchestrator, and image choices in full.

Deploying on Compute Engine (GCE)​

When deploying AI Hypercomputer clusters on Google Compute Engine, you have three main deployment types:

FactsGCE deployment types
  • Direct VM management
  • Orchestrated frameworks
  • Fully managed services

Additional resources​

0%0 of 147 pages studied