Why should you care?

CloudVerse fraud model retraining needs NC-series VMs or AKS GPU node pools.

See it live — copy this example

Use AWS/Azure/GCP free tier or local Docker/Kind. Sketches and YAML are meant to be typed and adapted.

# AKS GPU node pool (CloudVerse ML)
az aks nodepool add \
  --resource-group rg-cloudverse-ml \
  --cluster-name aks-cloudverse-ml \
  --name gpupool \
  --node-count 2 \
  --node-vm-size Standard_NC6s_v3 \
  --node-taints sku=gpu:NoSchedule \
  --labels workload=ml
# Pod spec requires toleration + nvidia.com/gpu resource

What happened?

• GPUs are expensive — schedule with taints, autoscale to zero where possible, and pin driver/CUDA versions.
• Follow the steps below — typing the code yourself is the fastest way to learn.

Practice next

1. Request GPU quota in region.
2. Add GPU node pool with taint.
3. Run one CUDA sample pod.
4. Use spot GPU for training.
5. Split inference onto CPU for small models.

Remember

Dedicated GPU pools. Taints control scheduling. Scale down when idle.

CloudVerse fraud retrain

Weekly model refresh on 2M transactions.

Outcome: GPU pool scales up for job, scales down after.