This chapter covers Committed Use Discounts (CUDs), a core cost optimization tool in Google Cloud that offers significant discounts in exchange for a 1- or 3-year commitment to specific resources. Understanding CUDs is critical for the ACE exam, as cost management questions appear in roughly 15-20% of the exam, and CUDs are a primary mechanism for reducing compute costs. This chapter will explain what CUDs are, how they work, how to purchase and manage them, and how they interact with other discount types like sustained use discounts and sole-tenant node discounts.
Jump to a section
Imagine a gym that charges per visit: $10 each time. You go 3 times a week, so you spend about $120 a month. The gym offers a membership: $80 a month for unlimited visits. If you commit to the membership, you save money even if you skip a week, but you pay even if you don't go. Now, the gym also has a 'flex plan' where you get 10% off per visit if you go at least twice a week, but no commitment. In Google Cloud, Committed Use Discounts (CUDs) work like the membership. You commit to a certain amount of vCPUs, memory, or other resources for 1 or 3 years, and you get a significant discount (up to 70%) compared to on-demand (pay-per-visit). The 'flex plan' is like sustained use discounts, which are automatic but smaller. Just like a gym membership, CUDs require you to pay for the committed resources regardless of usage, but they are the best way to save money if you have predictable workloads. The key is to match your commitment to your baseline usage—overcommitting wastes money, undercommitting leaves savings on the table.
What Are Committed Use Discounts (CUDs)?
Committed Use Discounts (CUDs) are a pricing model in Google Cloud that offers discounted prices for Compute Engine resources (vCPUs, memory, GPUs, and local SSDs) in exchange for a commitment to use a minimum amount of those resources for a 1-year or 3-year term. CUDs are region-specific and resource-type-specific. For example, you can commit to a certain number of vCPUs and memory in the us-central1 region. The discount is applied automatically to eligible usage within the region, regardless of which project or VM uses the resources, as long as they match the committed resource type and region.
CUDs are available for both general-purpose and optimized machine families, including N1, N2, N2D, E2, C2, C2D, C3, M1, M2, M3, and others. They also cover sole-tenant nodes, GPUs, and local SSDs. The discount percentage depends on the resource type and commitment term: 1-year commitments typically offer 20-40% discount, while 3-year commitments offer 40-70% discount compared to on-demand pricing.
How CUDs Work Internally
When you purchase a CUD, you are making a financial commitment to pay for a specified amount of resources each hour for the duration of the term. Google Cloud tracks your usage of the committed resources across all projects in the billing account. At the end of each hour, the system checks how much of the committed resources you used. If you used at least the committed amount, you pay the discounted rate for that amount. If you used less, you still pay the discounted rate for the committed amount (i.e., you pay for unused capacity). Any usage above the committed amount is charged at the standard on-demand rate, but may also be eligible for Sustained Use Discounts.
For example, suppose you commit to 10 vCPUs in us-central1 for 1 year. If in a given hour you use 12 vCPUs across all projects in that region, you pay the discounted rate for 10 vCPUs and the on-demand rate for the remaining 2 vCPUs. If you only use 8 vCPUs, you still pay the discounted rate for 10 vCPUs (the committed amount), and you effectively waste 2 vCPUs of capacity you paid for.
Key Components and Defaults
Commitment Term: 1 year or 3 years. 3-year commitments offer higher discounts.
Payment Options: Upfront (pay all at once) or monthly (pay each month). Upfront payments yield slightly higher discounts.
Resource Types: vCPUs, memory (in GB), GPUs (by model), local SSDs (in GB).
Machine Families: General-purpose (N1, N2, N2D, E2, C3) and optimized (C2, C2D, M1, M2, M3).
Regions: CUDs are per region. You cannot combine commitments across regions.
Projects: CUDs apply to all projects within the same billing account. You do not assign a CUD to a specific project.
Minimum Commitment: There is no minimum dollar amount, but you must commit to at least 1 vCPU or 1 GB of memory, etc.
Discount Calculation: Discounts are applied to the committed resource usage. The discount percentage is fixed at purchase time and does not change.
Purchasing and Managing CUDs
You can purchase CUDs through the Google Cloud Console, the gcloud command-line tool, or the API. The process involves specifying the region, resource type, amount, term, and payment option. Once purchased, the commitment is active immediately and cannot be canceled or refunded (except in very limited circumstances, such as when Google Cloud discontinues a machine series).
Example gcloud command to purchase a 1-year commitment for 10 vCPUs and 20 GB of memory in us-central1:
gcloud compute commitments create my-commitment \
--region=us-central1 \
--plan=12-month \
--resources=vcpu=10,memory=20GBTo list existing commitments:
gcloud compute commitments list --region=us-central1Interaction with Other Discounts
CUDs are applied before Sustained Use Discounts (SUDs). SUDs are automatic discounts that apply when you run a VM for a significant portion of the month. If you have a CUD, the committed usage is first charged at the discounted rate, and any additional usage beyond the commitment is eligible for SUDs. However, if your total usage is less than the commitment, you pay the committed amount and do not receive SUDs on the unused portion (because you are not actually using those resources).
CUDs also work with sole-tenant nodes. You can commit to sole-tenant nodes (e.g., sole-tenant-node-1), which gives you a discount on the node cost. The same rules apply: you commit to a certain number of nodes for 1 or 3 years.
Important Exam Considerations
CUDs are region-specific. A commitment in us-west1 does not apply to resources in us-east1.
CUDs are resource-type-specific. A commitment to N2 vCPUs does not apply to C2 vCPUs.
CUDs are not project-specific. They apply to all projects under the same billing account.
You can have multiple CUDs in the same region for different resource types or machine families.
If you need to change your commitment, you cannot modify an existing commitment; you must purchase a new one and let the old one expire. However, you can purchase overlapping commitments.
The discount is applied at the billing level, not at the VM level. You will see the discounted rate in your billing reports.
For GPUs, you must commit to a specific GPU model (e.g., NVIDIA Tesla T4).
Local SSDs are committed in GB increments.
Common Pitfalls and Exam Traps
Trap: CUDs apply to all regions. Wrong. They are per region.
Trap: CUDs can be shared across billing accounts. Wrong. They apply only to the billing account under which they were purchased.
Trap: You can cancel a CUD anytime. Wrong. CUDs are non-cancellable and non-refundable.
Trap: CUDs cover all machine types in a region. Wrong. They are specific to machine family and resource type.
Trap: 1-year commitments give the same discount as 3-year. Wrong. 3-year gives higher discount.
Trap: CUDs are applied to the VM's total cost, including OS licenses. Wrong. CUDs only cover the underlying compute resources (vCPU, memory, GPU, local SSD). Premium OS licenses (e.g., Windows, RHEL) are charged separately at on-demand rates.
Assess Baseline Resource Usage
Before purchasing a CUD, analyze your historical resource usage across all projects in the billing account. Use tools like the Google Cloud Billing reports, Recommender (which provides CUD recommendations), or third-party cost management tools. Identify the minimum amount of vCPUs, memory, GPUs, and local SSDs that you consistently use each hour. This baseline should be at least the amount you commit to. Overcommitting leads to waste; undercommitting leaves savings unrealized.
Choose Commitment Term and Payment
Decide between a 1-year or 3-year commitment. 3-year commitments offer higher discounts but lock you in longer. Also choose payment option: upfront (pay once) or monthly. Upfront payments provide a slightly higher discount. The choice depends on your cash flow and confidence in long-term resource needs. For the exam, remember that 3-year upfront gives the maximum discount.
Select Region and Resource Type
Determine which region your workloads run in. CUDs are region-specific, so you may need separate commitments for each region. Also specify the resource type: vCPUs, memory, GPUs (with model), or local SSDs. For memory, you commit to a specific amount in GB. For vCPUs, you commit to a number of vCPUs. Note that memory and vCPUs must be committed together in a single commitment if you want both; you cannot have separate commitments for vCPU and memory in the same region for the same machine family? Actually, you can create separate commitments for vCPUs and memory, but they are independent. However, to cover a full VM, you need both.
Purchase the Commitment
Use the Google Cloud Console, gcloud, or API to purchase the commitment. For example, using gcloud: `gcloud compute commitments create my-commitment --region=us-central1 --plan=12-month --resources=vcpu=10,memory=20GB`. Once purchased, the commitment starts immediately. You will see the discounted rate on your bill for resources that match the commitment. The commitment is non-cancellable, so double-check the details.
Monitor and Adjust Usage
After purchasing, monitor your usage to ensure you are meeting the commitment. Use billing reports and the Recommender to see utilization. If you consistently use more than committed, consider purchasing additional CUDs. If you use less, you may be wasting money. You cannot reduce a commitment, but you can let it expire and purchase a smaller one next time. Also, remember that CUDs are applied at the billing account level, so any project in that account benefits.
Enterprise Scenario 1: E-commerce Platform with Predictable Traffic
A large e-commerce company runs its production workloads on Compute Engine in us-central1. They have a baseline of 500 vCPUs and 2 TB of memory running 24/7 for their web servers and databases. Traffic spikes during holiday seasons, but the baseline is constant. They purchase a 3-year CUD for 500 vCPUs and 2 TB of memory in us-central1 with upfront payment, saving approximately 57% compared to on-demand. During spikes, additional resources are used on-demand. The CUD covers the baseline, and the on-demand costs for spikes are manageable. The company uses the Recommender to verify the baseline and sets up alerts if usage drops below 90% of commitment to avoid waste.
Enterprise Scenario 2: Machine Learning with GPU Workloads
A research lab runs machine learning training jobs using NVIDIA Tesla T4 GPUs in us-west1. They have a consistent need for 20 GPUs for training, plus occasional spikes for experimentation. They purchase a 1-year CUD for 20 T4 GPUs with monthly payments. The CUD gives them a 30% discount on the GPU cost. However, they forget to commit to the accompanying vCPUs and memory, so those are charged at on-demand rates. They later add a separate CUD for vCPUs and memory. The lab also uses sole-tenant nodes for compliance, and they purchase a CUD for those nodes as well.
Common Misconfiguration: Overcommitting
A startup purchases a 3-year CUD for 100 vCPUs based on projected growth, but their actual usage never exceeds 50 vCPUs. They end up paying for 50 unused vCPUs each hour for 3 years, wasting thousands of dollars. This could have been avoided by starting with a 1-year CUD for 50 vCPUs and then purchasing additional commitments as usage grows. The Recommender would have suggested a lower commitment based on historical data.
The ACE exam tests CUDs primarily under Objective 4.2: 'Manage and provision compute resources' and indirectly under cost management. Expect 2-3 questions on CUDs, often scenario-based where you must choose the most cost-effective option.
Most Common Wrong Answers and Why
'CUDs apply to all regions' – Candidates confuse CUDs with sustained use discounts, which are global. CUDs are per region. The exam will present a scenario with resources in multiple regions and ask for the best way to save costs. The correct answer is to purchase separate CUDs for each region.
'CUDs can be shared across billing accounts' – CUDs are tied to the billing account under which they are purchased. If an organization has multiple billing accounts, each needs its own CUDs. The exam may present a scenario where resources are in different billing accounts, and the candidate incorrectly selects a single CUD.
'1-year and 3-year commitments offer the same discount' – The discount percentage differs. 3-year commitments offer higher discounts (up to 70%) compared to 1-year (up to 40%). The exam may ask which commitment yields the highest discount; the answer is 3-year upfront.
'CUDs cover all machine families in a region' – CUDs are specific to a machine family (e.g., N2, C2). A commitment to N2 vCPUs does not cover C2 vCPUs. The exam may include a scenario where a company uses both N2 and C2 instances and asks for the correct CUD strategy.
Specific Numbers and Terms to Memorize
Discount ranges: 1-year: 20-40%; 3-year: 40-70%.
Commitment terms: 12-month or 36-month.
Payment options: Upfront (full payment) or monthly.
Resource types: vCPU, memory (GB), GPU (by model), local SSD (GB).
Region-specific: Yes.
Non-cancellable: Yes.
Applies to all projects in the billing account: Yes.
Edge Cases and Exceptions
CUDs for sole-tenant nodes are available.
CUDs for GPUs require specifying the GPU model.
If Google Cloud discontinues a machine series, you may be able to cancel the CUD without penalty.
CUDs do not cover premium OS licenses (e.g., Windows, RHEL) or additional services like networking or disks.
How to Eliminate Wrong Answers
If the question mentions multiple regions, eliminate any answer that suggests a single CUD.
If the question mentions different machine families, eliminate answers that treat them as interchangeable.
If the question asks for maximum savings, look for 3-year upfront.
If the question asks about flexibility, 1-year monthly is more flexible but lower discount.
CUDs require a 1-year or 3-year commitment to specific resources in a specific region.
CUDs are non-cancellable and non-refundable.
CUDs apply to all projects under the same billing account.
3-year commitments offer higher discounts than 1-year.
Upfront payment yields slightly higher discount than monthly.
CUDs are resource-type-specific (e.g., N2 vCPUs, T4 GPUs).
CUDs do not cover premium OS licenses or additional services.
These come up on the exam all the time. Here's how to tell them apart.
Committed Use Discounts (CUDs)
Requires a 1- or 3-year commitment.
Offers higher discounts (up to 70%).
Region-specific and resource-type-specific.
Non-cancellable; you pay even if unused.
Applied before SUDs.
Sustained Use Discounts (SUDs)
Automatic, no commitment required.
Offers lower discounts (up to 30% for running full month).
Applied globally across all regions.
Only applies to actual usage; no penalty for low usage.
Applied after CUDs.
Mistake
CUDs apply to all resources in a project automatically.
Correct
CUDs apply to all projects under the same billing account, not just a single project. They are not project-specific.
Mistake
You can cancel a CUD at any time if you no longer need the resources.
Correct
CUDs are non-cancellable and non-refundable. Once purchased, you pay for the entire term regardless of usage.
Mistake
CUDs cover all machine types, including GPUs and local SSDs, without specifying the model.
Correct
For GPUs, you must commit to a specific GPU model (e.g., T4, V100). Local SSDs are committed in GB. The commitment is resource-type-specific.
Mistake
A 1-year CUD gives the same discount as a 3-year CUD.
Correct
3-year commitments offer higher discounts (up to 70%) compared to 1-year (up to 40%). The discount percentage increases with longer terms.
Mistake
CUDs are automatically applied to the most expensive resources first.
Correct
CUDs are applied to any eligible usage in the region, regardless of cost. The discount is applied to the committed amount, not prioritized by cost.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
No, CUDs are per region. You must purchase separate CUDs for each region where you want discounted resources. For example, if you have workloads in `us-central1` and `us-west1`, you need two CUDs.
You still pay the discounted rate for the full committed amount. For example, if you commit to 10 vCPUs but only use 8, you pay for 10 at the discounted rate. This is why it's important to commit only to your baseline usage.
No, CUDs are non-cancellable and non-refundable. You cannot modify the commitment amount or term. However, you can purchase additional CUDs if you need more capacity.
No, CUDs only cover the underlying compute resources: vCPUs, memory, GPUs, and local SSDs. Networking, persistent disks, and premium OS licenses (e.g., Windows) are billed separately at on-demand rates.
CUDs are applied first. Any usage beyond the committed amount is eligible for Sustained Use Discounts. If your total usage is less than the commitment, you do not receive SUDs on the unused portion because you are paying for it anyway.
Yes, you can commit to sole-tenant nodes. The commitment is for the number of nodes, and you get a discount on the node cost. The same rules apply: region-specific, non-cancellable, etc.
Use the Google Cloud Recommender, which analyzes your historical usage and provides CUD recommendations. You can also use billing reports to identify your baseline usage. It's safer to start with a lower commitment and add more later.
You've just covered Committed Use Discounts (CUDs) — now see how well it sticks with free ACE practice questions. Full explanations included, no account needed.
Done with this chapter?