GCP Sustained & Committed Use Discount Calculator
Model sustained-use and committed-use discounts together. SUD applies per quarter of the month and is automatic; a CUD covers usage at the committed rate and forfeits the SUD it would have earned — so the saving is the difference, not the headline.
The three prices
| List price7,300 instance-hours | $1,387.00 |
| With sustained use20.0% off, automatic | $1,109.88 |
| With a commitmentChoose one on the left | — |
| You would pay | $1,109.88 |
Why half a month does not earn half the discount
Google discounts each quarter of the month at its own rate rather than applying one rate to the total. On the 20% band the four quarters bill at 100%, 87%, 73%, 60% of on-demand, so a full month averages 20% off — but half a month averages only the first two, which is 6.6%.
The curve is convex, so the last quarter of uptime is worth far more than the first. That is also why scheduling a machine off overnight costs more discount than the hours suggest.
What this model assumes
- The on-demand rate is an input, not a lookup. Google’s machine-type price table is rendered in the browser and returns no rates to any public fetch, and the Cloud Billing Catalog API needs a key. Rather than type rates from memory, this asks for the one number you can read off the pricing calculator in ten seconds. The same choice is made for GKE and AKS node prices elsewhere on this site.
- Sustained-use discounts apply per quarter of the month, per resource, and are automatic. The machine-family bands are read from the English documentation: N2, N2D and C2 cap at 20%; N1 and memory-optimized cap at 30%. The translated version of that page orders the columns the other way round.
- Committed and sustained use do not stack. Covered usage bills at the committed rate and forfeits its sustained-use discount, so the saving shown is always measured against sustained use rather than against list.
- Compute flexible CUD rates are verified (28% for one year, 46% for three) and apply across Compute Engine, GKE and Cloud Run. Resource-based rates are published only as documented maxima and are Compute Engine only. A one-year resource-based term is not offered here because Google publishes no single percentage for it.
- Spot and preemptible instances, sole-tenant nodes, licence commitments and regional commitment sharing are not modelled.
Frequently asked questions
Do committed use discounts and sustained use discounts stack?
No, and this is the most expensive misunderstanding in Google Cloud commitment planning. Usage covered by a commitment bills at the committed rate and earns no sustained-use discount. So the value of a commitment is the difference between the two discounts, not the commitment percentage. A workload already running all month on an N1 machine earns a 30% sustained-use discount automatically; buying a 28% one-year flexible commitment for it is worse than doing nothing.
How does the Google Cloud sustained use discount actually work?
It is applied per quarter of the month, not as a flat rate on the total. Each quarter a resource runs is discounted at its own multiplier, and the discount you see is the average of the quarters you used. On the 30% band the four quarters bill at 100%, 80%, 60% and 40% of on-demand — so a full month averages 30% off, but half a month averages only the first two quarters, which is 10%. The curve is convex, so the final quarter of uptime is worth far more discount than the first.
Which machine types get a 30% sustained use discount?
N1 general-purpose machine types (predefined and custom), N1 sole-tenant nodes, and memory-optimized machine types reach 30%. N2, N2D and C2 — including their sole-tenant nodes — cap at 20%. It is worth checking against the English documentation: the translated version of that page orders the two columns differently, and reading it there gives exactly the wrong mapping.
What is the difference between a flexible and a resource-based CUD?
A compute flexible CUD is spend-based: you commit to an hourly dollar amount and the discount applies across Compute Engine, GKE and Cloud Run regardless of machine type or region. Published rates are 28% for one year and 46% for three. A resource-based CUD is Compute Engine only and commits you to specific vCPU and memory in a specific region, in exchange for a deeper discount — documented as up to 55%, or up to 70% for memory-optimized series.
Why does this calculator ask for the on-demand price instead of looking it up?
Because Google does not publish one that can be read. The machine-type rate table on the Compute Engine pricing page is rendered in the browser and returns no rates to any fetch, the old public price list JSON was retired, and the Cloud Billing Catalog API requires an API key. The honest options were to ask you for one number or to type rates from memory, and typing rates from memory is exactly the failure this site exists to correct. The same choice is made for GKE and AKS node prices in the Kubernetes calculator.
The hard part is knowing which workloads are already discounted.
Finitizer reads your Compute Engine usage hour by hour, works out the sustained-use discount each workload already earns, and sizes commitments against what is left — so you do not buy a commitment that replaces a bigger discount with a smaller one.