How do I choose CPU values for Kubernetes?

Measure real usage, set requests near steady-state (P90), and skip CPU limits unless you need strict caps to avoid throttling.

Should Kubernetes memory be in MiB or MB?

Use binary units (Mi/Gi) because Kubernetes treats memory as powers of two. Cloud consoles show decimal GB/MB, so convert to compare correctly.

What is a millicore in Kubernetes?

1000 millicores equal one CPU core. A request of 500m is half a core; limits above the request can cause throttling if exceeded.

← Back to Tools

Kubernetes Resource Units Converter

Convert CPU and memory units fast. Keep manifests consistent, avoid decimal/binary mixups, and snap memory to clean base-2 values.

CPU

1 core = 1000m

Value Unit

Cores 0

Millicores 0

NanoCPUs 0

Memory

Binary (Mi/Gi) recommended

Value Unit

Bytes 0

MiB (binary) 0

GiB (binary) 0

GB (decimal) 0

Nearest power-of-two (MiB) 0

How to use

Enter a value and choose its unit.
Read the converted numbers instantly—no refresh or copy buttons.
Use power-of-two MiB for clean node packing and predictable limits.

Sizing shortcuts

Requests: steady-state; Limits: bursts (skip CPU limits if you can).
Memory: prefer 512Mi, 1024Mi, 2048Mi, 4096Mi, etc. for fewer surprises.
Match pod totals to node allocatable to avoid fragmentation.

What is a millicore?

Kubernetes measures CPU in cores and millicores. 1000m = 1 core. A request of 500m means the pod needs half a core; a limit of 2000m caps the pod at two cores.

Schedulers place pods based on requests, not limits. If you set a high limit but a tiny request, you risk noisy-neighbor throttling later. Keep requests close to realistic steady-state.

CPU requests vs limits in Kubernetes

Requests

Used by the scheduler to fit pods on nodes.
Guarantees the requested CPU is available.
Set to steady-state usage; measure with metrics first.

Limits

Caps CPU with CFS quotas; exceeding it triggers throttling.
Optional—many teams skip CPU limits to avoid latency spikes.
Keep limits close to requests if you must set them.

If you do set CPU limits, pair them with sensible requests. Use Horizontal Pod Autoscalers to scale out instead of depending on large CPU bursts.

Binary vs decimal memory units

Kubernetes treats Mi/Gi as powers of two (1 Gi = 1024 Mi). Cloud consoles often show decimal GB (1 GB = 1000 MB). Always convert before comparing usage to cloud bills to avoid false overages.

Base-2 rounding matters for packing. Sizes like 512Mi, 1024Mi, 2048Mi reduce fragmentation and align with typical node capacities. The converter shows the nearest power-of-two MiB automatically.

Fast FAQ

How do I choose CPU values?

Measure real usage, set requests near P90 steady-state, and skip limits unless you need strict caps.

What about memory?

Memory limits are hard: exceeding them OOM-kills the pod. Keep limits slightly above observed peaks and use base-2 values for clarity.

Does this work for Helm values?

Yes. Paste the converted numbers into resources.requests and resources.limits in your chart values.