All tools
Free architecture tool

GCP Compute Selector

Cloud Run, GKE, Cloud Functions or Compute Engine? Answer six quick questions and get a senior-level recommendation — with the reasoning, the alternatives and when NOT to use it.

No sign-up. Nothing leaves your browser. Instant answer.

Questions answered: 0 / 6
  1. 1What are you running?
  2. 2When there’s no traffic?
  3. 3Do you need Kubernetes?
  4. 4Special hardware?
  5. 5How much ops do you want to own?
  6. 6Runtime needs?

How I decide in the real world

  • Start at the most managed tier that fits, then only "graduate" to more control when a workload truly needs it.
  • Scale-to-zero beats a warm cluster for spiky or low-traffic work — idle nodes are pure waste.
  • Kubernetes is a platform, not a default. Adopt it when you need its API, not because it’s on the CV.
  • Match ops cost to team maturity — the cheapest compute is the one your team can actually operate.

Common questions

Cloud Run or GKE?

Cloud Run is the right default for stateless containers and HTTP/event workloads — it scales to zero and there is no cluster to run. Reach for GKE when you need the Kubernetes API itself: DaemonSets, operators, service meshes, GPUs, or workloads that never scale to zero. Most teams start on Cloud Run and only graduate to GKE when a concrete need appears.

When should I use Cloud Functions instead of Cloud Run?

Both are serverless. Cloud Functions fits small, single-purpose event handlers where you just want to ship a function. Cloud Run gives you a full container, more control over runtime, concurrency and dependencies, and handles heavier HTTP services. When in doubt, Cloud Run is the more flexible default — it can do almost everything Functions can, plus more.

GKE Autopilot or Standard?

Autopilot is the default: Google manages nodes, and you pay per pod, which removes most of the operational burden. Choose Standard only when you need node-level control — custom machine types, GPUs/TPUs, privileged DaemonSets or specific kernel settings. Start on Autopilot and switch to Standard when a workload genuinely requires it.

When is Compute Engine still the right choice?

Compute Engine (raw VMs) fits legacy or stateful software that is not containerised, workloads needing a specific OS/kernel, licensed software tied to a VM, or lift-and-shift migrations. If you can containerise, a managed tier is usually cheaper to operate — but VMs remain the honest answer when the workload demands full machine control.

Guidance based on 2026 Google Cloud capabilities and real-world platform experience — not a substitute for a full architecture review. Uses no cookies and sends no data server-side.