What is a Millicore?

A millicore (abbreviated m) is Kubernetes' unit for measuring CPU allocation. It's 1/1000th of a CPU core.

Think of it like currency. If a vCPU is a dollar, a millicore is a tenth of a cent. This granularity lets Kubernetes pack workloads efficiently — a small web server might need only 100m (10% of a core), while a data preprocessing job might need 4000m (4 full cores).

# Kubernetes pod spec
resources:
  requests:
    cpu: "4000m"      # 4 vCPUs
    memory: "16Gi"
    nvidia.com/gpu: 1  # 1 GPU
  limits:
    cpu: "8000m"      # burst up to 8 vCPUs
    memory: "32Gi"
    nvidia.com/gpu: 1

Many GPU cloud providers (CoreWeave, RunPod, Lambda) run Kubernetes under the hood. When you deploy a GPU workload as a container, you specify CPU in millicores alongside your GPU request.

Getting this wrong has real consequences:

• Too few millicores → CPU can't feed data to the GPU fast enough → GPU sits idle → you're paying for unused compute
• Too many millicores → you're reserving CPU you don't need → higher cost, potentially blocking other workloads

GIS uses compute.vcpus (whole vCPUs), not millicores. To convert:

millicores = compute.vcpus × 1000

// Lambda H100: 26 vCPUs = 26,000m available
// AWS p5.48xlarge: 192 vCPUs = 192,000m available

When deploying on Kubernetes-based providers, you'll request a fraction of the instance's total vCPUs in millicores for your pod.