k3s / Kubernetes Clusters with Integrated GPUs, No €500/Month Nodes
Standing up a Kubernetes cluster with GPUs doesn’t mean renting nodes at €300-500/month each. With servers carrying an integrated GPU (iGPU / APU) exposed through a device plugin, your pods request the GPU like any standard resource — from just €14.99/month per node.
One GPU per node, exposed to the cluster
The iGPU/APU is presented to Kubernetes through a GPU device plugin: your
Deployments reserve the GPU like any other resource, whether on lightweight
k3s or full k8s. Perfect for a staging, CI or inference cluster without
blowing the budget.
| Requirement | Dedicated-GPU nodes (~€300-500/month) | Linux GPU iGPU nodes (from €14.99/month) |
|---|---|---|
| Staging / GPU CI cluster | ✅ expensive | ✅ affordable |
| Containerized inference / media | ✅ wasted | ✅ well suited |
| Adding more nodes | 💸💸💸 | 💶 |
Reserving the GPU in a pod
resources:
limits:
gpu.intel.com/i915: '1'
The pod gets access to the node’s iGPU. The scheduler places the workload wherever a GPU is available.
HA, GitOps and access
- Multiple nodes, GPU scheduling, Ingress / LoadBalancer;
- GitOps with ArgoCD or Flux for declarative deployments;
- Root SSH + kubeconfig provided;
- k3s cluster option +5 €/node, managed node +5 €/node (monitoring, updates and security handled by us).
FAQ
Which workloads run well on an iGPU cluster? AI inference (OpenVINO), video transcoding (VAAPI), headless rendering, browser testing: the GPU-light workloads, spread across multiple nodes.
k3s or full k8s? k3s for lightness and simplicity, full k8s if you already have the tooling. Both expose the GPU the same way.
Can I do GitOps from day one? Yes: ArgoCD/Flux point at your repository and apply the desired state.
Conclusion
For a GPU cluster for staging, CI or inference, integrated-GPU nodes get the job done without the bill of dedicated-GPU nodes. Start from €14.99/month per node, +5 €/node for the k3s cluster option.