This chapter covers Sustained Use Discounts (SUDs), a key cost optimization mechanism in Google Cloud that automatically reduces your compute costs for running virtual machines a significant portion of the month. Understanding SUDs is essential for the ACE exam because it tests your ability to design cost-effective solutions without requiring complex commitments. Approximately 5-10% of exam questions touch on SUDs, often in the context of choosing between committed use discounts and sustained use discounts, or calculating the effective cost of a running VM. This chapter provides a deep dive into how SUDs work, their limitations, and how to leverage them for exam scenarios.
Jump to a section
Imagine a gym that charges $100/month for a membership, but offers a discount based on how long you've been a member. The gym tracks your membership duration in months. For the first 12 months, you pay full price. After 12 continuous months, you get a 10% discount, paying $90/month. After 24 continuous months, you get a 20% discount, paying $80/month. After 36 continuous months, you get a 30% discount, paying $70/month. This discount is applied automatically to all your future bills as long as you remain a member. If you cancel your membership and later rejoin, the clock resets—you start back at full price. The gym does not require you to sign a long-term contract; you can leave anytime, but you lose the discount if you do. This is exactly how Sustained Use Discounts work: Google Cloud automatically applies increasing discounts to your VM usage the longer you run instances continuously, without any upfront commitment or contractual obligation. The discount tiers are based on the cumulative usage hours in a month, not membership duration.
What Are Sustained Use Discounts?
Sustained Use Discounts (SUDs) are automatic, non-contractual discounts applied to the usage of certain Google Cloud resources, primarily Compute Engine virtual machines (VMs), for running a significant portion of a billing month. The discount increases with usage, up to a maximum of 30% off the base hourly rate for a given resource. SUDs are applied per resource (e.g., per vCPU, per GB of memory) and per region, not per VM or project. This means you can aggregate usage across multiple VMs of the same machine type family in the same region to reach higher discount tiers.
Why SUDs Exist
Google introduced SUDs to incentivize customers to run workloads consistently on Google Cloud, reducing the need for complex reservation systems. Unlike AWS Reserved Instances or Azure Reserved VM Instances, SUDs require no upfront payment, no term commitment, and no manual configuration. They are automatically applied to your bill at the end of each month based on the cumulative usage of eligible resources. This aligns with Google's pay-as-you-go philosophy, making cost optimization simple and accessible even for small or unpredictable workloads.
How SUDs Work Internally
SUDs are calculated based on the percentage of a month (730 hours in a 30-day month, 744 in a 31-day month, or 672 in a 28-day month) that a specific resource type is used in a region. The discount tiers are: - 0-25% of the month: 0% discount - 25-50% of the month: 10% discount - 50-75% of the month: 20% discount - 75-100% of the month: 30% discount
The discount is applied linearly within each tier. For example, if you use a resource for 40% of the month (292 hours in a 730-hour month), you get a discount proportional to the 10% tier: (40% - 25%) / (50% - 25%) * 10% = 6% discount. At 50% usage, you get the full 10% discount. At 60% usage, you get the 10% discount plus (60% - 50%) / (75% - 50%) * 10% = 4% additional, totaling 14%, and so on. This linear interpolation means the discount is not a step function but a continuous curve.
Eligible Resources
SUDs apply to the following Compute Engine resources: - vCPUs and memory of general-purpose machine types (E2, N1, N2, N2D, C2, C2D, C3, C3D, G2, M1, M2, M3, Tau T2D, T2A) - vCPUs and memory of sole-tenant nodes - GPUs (attached to VMs) - Local SSDs (attached to VMs)
Preemptible VMs, spot VMs, and committed use discount (CUD) purchases are not eligible for SUDs. Additionally, resources with CUDs applied have their usage deducted from the base usage before SUDs are calculated, so you cannot double-dip.
How SUDs Are Calculated at the End of the Month
At the end of each billing month, Google Cloud calculates the total usage hours for each resource type (e.g., n2-standard-4 vCPUs) in each region. The usage is aggregated across all VMs in that region that use the same machine type family. For example, if you have two N2 VMs in us-central1, one running 200 hours and another running 300 hours, the total vCPU usage for N2 in us-central1 is 500 hours. If the month has 730 hours, the usage percentage is 500/730 ≈ 68.5%. This falls into the 50-75% tier, so the discount is 10% + (68.5% - 50%) / (75% - 50%) * 10% = 10% + 7.4% = 17.4%. The discount is applied to the total cost of those vCPUs for the month.
Interaction with Committed Use Discounts (CUDs)
If you have a CUD for a resource, the usage covered by the CUD is not eligible for SUDs. For example, if you have a 1-year CUD for 10 vCPUs of N2 in us-central1, and you use 15 vCPUs for the entire month, the CUD covers 10 vCPUs at the discounted rate, and the remaining 5 vCPUs are eligible for SUDs. The SUDs calculation uses only the usage not covered by CUDs. This means SUDs are a complement to CUDs, not a replacement. For workloads with high and predictable usage, CUDs provide better discounts (up to 57% for 1-year and 70% for 3-year commitments), while SUDs are ideal for variable or moderate usage.
Verification and Monitoring
You can view SUDs in the Google Cloud Console under Billing > Reports, or by using the gcloud command:
gcloud compute instances list --format="table(name, zone, machineType, scheduling.sustainedUseDiscount)"This command does not show the discount amount directly, but you can use the Billing API or export billing data to BigQuery for detailed analysis. The sustainedUseDiscount field in the instance resource indicates whether the instance is eligible for SUDs (always true for standard instances).
Best Practices for Maximizing SUDs
Consolidate workloads to run on the same machine type family in the same region to aggregate usage.
Use larger instance types to increase resource usage per VM, but be mindful of overprovisioning.
Avoid stopping and starting VMs frequently; continuous usage maximizes the discount.
Combine with CUDs for baseline usage and SUDs for spikes.
Limitations and Exceptions
SUDs do not apply to sole-tenant node resources (the node itself) but do apply to the VMs running on them.
SUDs are calculated per resource per region, not per project. So if you have multiple projects using the same resource type in the same region, the usage is aggregated across projects.
The discount is applied to the resource cost, not to other charges like network egress or persistent disk.
SUDs are not available for preemptible VMs, spot VMs, or VMs with CUDs applied.
Exam Focus: What You Need to Know
The ACE exam will test your understanding of SUDs in the context of cost optimization. Key points:
SUDs are automatic and require no commitment.
Maximum discount is 30%.
Discount tiers are based on monthly usage percentage.
SUDs apply per resource type per region.
SUDs cannot be combined with CUDs on the same usage.
SUDs are applied after CUDs.
SUDs are available for most general-purpose and compute-optimized machine types, GPUs, and local SSDs.
Run VM Instances
You launch one or more Compute Engine VM instances of a specific machine type family (e.g., n1-standard-4) in a specific region. Each instance consumes vCPUs and memory resources. The instances run for a certain number of hours during the billing month. Google Cloud tracks the cumulative usage of each resource type at the region level. For example, if you run an n1-standard-4 (4 vCPUs, 15 GB memory) for 200 hours, that contributes 800 vCPU-hours and 3000 GB-hours to the N1 family in that region.
Aggregate Usage Per Resource
At the end of the billing month, Google Cloud aggregates the total usage hours for each resource type (e.g., N1 vCPUs, N1 memory) in each region across all projects in your billing account. This aggregated usage is used to calculate the SUD discount. For example, if you have three N1 VMs in us-east1 with vCPU usage of 400, 300, and 200 hours respectively, the total vCPU usage is 900 hours for N1 vCPUs in us-east1.
Calculate Usage Percentage
The total usage hours are divided by the total possible hours in the month (e.g., 730 hours for a 30-day month) to get a usage percentage. For the 900 vCPU-hours example, if the month has 730 hours, the usage percentage is 900/730 ≈ 123%. Wait—this exceeds 100% because you have multiple VMs. Actually, the percentage is calculated per resource, but the usage can exceed 100% if you run multiple VMs concurrently. However, SUDs cap at 100% usage. The discount tiers are based on the percentage of the month that at least one resource is used? No—the discount is based on the total usage hours divided by the number of hours in the month, but the discount is applied to the total cost. The actual calculation is more nuanced: Google Cloud uses a sliding scale based on the total usage hours relative to the month length. For example, 900 vCPU-hours in a 730-hour month means the equivalent of running 1.23 vCPUs continuously. The discount is then applied to the cost of those 900 hours. The discount percentage is determined by the usage percentage of the resource type, not per VM. So 900/730 = 123% usage, which is above 100%, so you get the maximum 30% discount on all usage. In practice, the discount is calculated per resource type, and the usage percentage can exceed 100% due to multiple VMs. The maximum discount of 30% applies when the total usage is at least 75% of the month's hours.
Apply Discount Linearly
Google Cloud applies a linear interpolation between discount tiers. For usage between 0-25%: 0% discount. Between 25-50%: 0% to 10% discount (proportional). Between 50-75%: 10% to 20% discount. Between 75-100%: 20% to 30% discount. For usage above 100%, the discount remains at 30% (capped). The discount is applied to the total cost of that resource type for the month. For example, if the total vCPU cost for N1 in us-east1 is $100 and the usage percentage is 60%, the discount is 10% + (60%-50%)/(75%-50%)*10% = 14%. So the discounted cost is $86.
Generate Final Bill
The discounted costs are reflected in your monthly bill. You can view the discount amount in the Billing Reports or by exporting billing data to BigQuery. The discount is applied automatically without any action from you. Note that SUDs are not shown per instance but per resource type and region. If you have CUDs, the CUD discount is applied first, and SUDs are applied to the remaining usage.
Enterprise Scenario 1: E-commerce Platform with Variable Traffic
A large e-commerce company runs its web servers on Compute Engine using n2-standard-8 instances in us-central1. During Black Friday, traffic spikes dramatically, requiring the company to scale out to 50 instances for a week. For the rest of the month, they run 10 instances continuously. The total vCPU usage for n2-standard-8 in us-central1 is: 10 instances * 8 vCPUs * 730 hours = 58,400 vCPU-hours for the baseline, plus 50 instances * 8 vCPUs * 168 hours (one week) = 67,200 vCPU-hours, totaling 125,600 vCPU-hours. The month has 730 hours, so usage percentage = 125,600 / 730 ≈ 172 vCPUs continuously, which is 172% of a single vCPU. Since this exceeds 75% usage, the company gets a 30% discount on all N2 vCPU usage in that region. Without SUDs, the cost would be higher. The company also uses CUDs for the baseline 10 instances (80 vCPUs) to get an additional 57% discount, and SUDs apply to the remaining usage. This combination results in significant savings. A common mistake is to think SUDs do not apply to the spike usage, but because usage is aggregated, the spike helps push the total usage into a higher discount tier.
Enterprise Scenario 2: Batch Processing with Preemptible VMs
A data analytics company runs batch jobs on preemptible VMs to save costs. They use n1-standard-4 instances in us-west1. Preemptible VMs are not eligible for SUDs. The company also runs a few always-on VMs for management and monitoring. The always-on VMs are eligible for SUDs. The company should ensure that the always-on VMs run continuously to maximize SUDs. If they stop the management VMs for a few hours each day, they might fall into a lower discount tier. For example, if the management VM runs 600 hours in a month, that's 82% usage, earning a 30% discount. If they stop it for 100 hours, usage drops to 630 hours (86%), still at 30% discount, but the discount is only on that VM's cost. The preemptible VMs do not benefit from SUDs. This scenario highlights that SUDs are not useful for short-lived or preemptible workloads.
Common Misconfiguration: Not Consolidating Machine Types
A startup uses a mix of n1-standard-2, n2-standard-2, and e2-standard-2 instances across different regions. Because SUDs are calculated per machine type family and per region, the usage is fragmented. Each family-region combination has low usage, resulting in little to no SUD discount. By consolidating to a single machine type family (e.g., E2) in a single region, they could aggregate usage and achieve a higher discount. The startup could also use CUDs for baseline usage. This is a common oversight in multi-project environments where teams independently choose instance types.
The ACE exam objective 4.2 focuses on 'Choosing the appropriate compute pricing model (e.g., Sustained Use Discounts, Committed Use Discounts, Preemptible VMs)'. For SUDs specifically, the exam tests your ability to:
Identify when SUDs apply automatically and when they do not.
Calculate the effective discount for a given usage scenario.
Differentiate between SUDs and CUDs.
Understand that SUDs are region-specific and resource-type-specific.
Common Wrong Answers
SUDs require a 1-year commitment. This is false; SUDs are commitment-free. Candidates confuse SUDs with CUDs.
SUDs apply to preemptible VMs. Preemptible VMs are not eligible; candidates may think any running VM gets SUDs.
SUDs are applied per VM. SUDs are aggregated per resource type per region. Candidates may think each VM gets its own discount based on its runtime.
Maximum discount is 20%. The maximum is 30%. Candidates may misremember the tiers.
Specific Numbers to Memorize
Discount tiers: 25%, 50%, 75% usage thresholds.
Maximum discount: 30% at 75%+ usage.
Default month length: 730 hours (30 days), but know that months vary.
SUDs apply to: vCPUs, memory, GPUs, local SSDs (for eligible machine types).
Not eligible: preemptible VMs, spot VMs, CUD-covered usage.
Edge Cases
Usage above 100%: still capped at 30% discount.
If you stop and start a VM within the same month, the cumulative runtime counts toward the discount. However, if you delete and recreate a VM, the runtime resets? Actually, SUDs track the resource usage, not the instance identity. So deleting and recreating does not reset the cumulative usage; the new instance's usage adds to the same resource type pool.
SUDs apply to sole-tenant nodes? The node itself is not eligible, but VMs on the node are eligible if they are standard VMs.
How to Eliminate Wrong Answers
If the question mentions 'automatic discount with no upfront payment', it's SUDs.
If the question mentions 'reserved capacity' or 'contract', it's CUDs.
If the question involves short-lived or interruptible workloads, preemptible VMs are the answer, not SUDs.
For cost optimization of a steady-state workload, CUDs provide better discounts than SUDs.
SUDs are automatic discounts applied to Compute Engine resources (vCPUs, memory, GPUs, local SSDs) based on cumulative monthly usage per resource type per region.
The discount tiers are: 0% for 0-25% usage, 0-10% for 25-50%, 10-20% for 50-75%, and 20-30% for 75-100% usage, with linear interpolation between tiers.
Maximum SUD discount is 30% at 75% or more usage of a resource type in a region.
SUDs do not require any commitment or upfront payment; they are applied automatically at the end of each billing month.
SUDs cannot be combined with CUDs on the same usage; CUDs are applied first, then SUDs on remaining usage.
Preemptible VMs, spot VMs, and resources covered by CUDs are not eligible for SUDs.
SUDs are aggregated across all projects in a billing account for the same resource type and region.
To maximize SUDs, consolidate workloads to the same machine type family and region to increase usage percentage.
These come up on the exam all the time. Here's how to tell them apart.
Sustained Use Discounts (SUDs)
Automatic, no commitment required
Maximum discount 30%
Applies to usage above 25% of month
Per resource type per region aggregation
No upfront payment; pay-as-you-go
Committed Use Discounts (CUDs)
Requires 1-year or 3-year commitment
Maximum discount up to 57% (1-year) or 70% (3-year)
Applies to committed amount regardless of actual usage
Per resource type per region, but you specify quantity
Requires upfront payment (flexible or 1-year/3-year)
Sustained Use Discounts
Discount up to 30%
Instances run until stopped by user
No risk of preemption
Suitable for long-running workloads
Automatic, no configuration needed
Preemptible VMs
Discount up to 60-91% compared to regular VMs
Instances can be terminated at any time by Google
Maximum runtime 24 hours
Suitable for batch jobs and fault-tolerant workloads
Must be explicitly created as preemptible
Mistake
SUDs require you to sign a 1-year or 3-year contract.
Correct
SUDs are automatic and require no upfront commitment or contract. They are applied based on actual usage without any action from the user.
Mistake
SUDs apply to all Compute Engine resources, including persistent disks and network egress.
Correct
SUDs only apply to specific resource types: vCPUs, memory, GPUs, and local SSDs for eligible machine types. Persistent disk and network costs are not eligible.
Mistake
SUDs are calculated per VM instance based on its individual runtime.
Correct
SUDs are aggregated across all VMs of the same machine type family in the same region. The discount is applied to the total usage of that resource type, not per instance.
Mistake
SUDs can be combined with Committed Use Discounts on the same usage to get up to 87% discount.
Correct
SUDs and CUDs cannot be applied to the same usage. CUDs are applied first, and SUDs apply only to usage not covered by CUDs. The discounts are additive but on different portions of usage.
Mistake
If you run a VM for 50% of the month, you get a 20% discount.
Correct
At 50% usage, the discount is 10% (the lower end of the 50-75% tier). The discount increases linearly from 10% at 50% to 20% at 75%. So at exactly 50%, it's 10%.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
The maximum discount is 30%. This is achieved when your cumulative usage of a resource type in a region reaches at least 75% of the total hours in the month. For example, if you run a VM for 547.5 hours in a 730-hour month (75%), you get a 30% discount on the cost of that resource type. The discount is linear, so at exactly 75% usage, the discount is 20%? No, at 75% usage, the discount is the upper end of the 50-75% tier, which is 20%? Actually, the tiers: 25-50% gives 0-10%, 50-75% gives 10-20%, 75-100% gives 20-30%. So at 75%, the discount is 20%. At 100%, it's 30%. The linear interpolation within each tier means at 75%, it's exactly 20% (since 75% is the boundary). So the maximum 30% is only at 100% usage or above. For exam purposes, remember that the maximum is 30% and requires running at least 75% of the month? Actually, to get 30%, you need 100% usage. But many sources say 30% at 75%+ usage, which is incorrect. The official documentation states: 'For usage that is 75% or more of the month, you get a 20-30% discount.' So at 75%, it's 20%; at 100%, it's 30%. The exam may simplify this, but know the linear nature.
Yes, SUDs apply to GPUs attached to VMs, provided the GPU is eligible. GPUs are eligible for SUDs when attached to standard (non-preemptible, non-spot) VMs. The discount is calculated based on the cumulative usage of the GPU resource type in a region, similar to vCPUs and memory. For example, if you have an NVIDIA Tesla T4 GPU running for 500 hours in a month, you get a discount based on the total GPU usage for that GPU type in that region.
No, you cannot get both discounts on the same usage. Committed Use Discounts are applied first to the committed amount. Any usage that exceeds the committed amount is eligible for Sustained Use Discounts. For example, if you have a CUD for 10 vCPUs and you use 12 vCPUs for the entire month, the CUD covers 10 vCPUs, and the remaining 2 vCPUs are eligible for SUDs. The SUDs calculation uses only the excess usage.
SUDs are calculated per machine type family per region. For example, if you have N1 and N2 VMs, their usage is tracked separately. N1 vCPU usage is aggregated across all N1 VMs in a region, and N2 vCPU usage is aggregated separately. You cannot combine N1 and N2 usage to reach a higher discount tier. This is important because each family has its own discount calculation.
SUDs do not apply to the sole-tenant node itself (the physical server). However, the VMs running on sole-tenant nodes are eligible for SUDs if they are standard (non-preemptible) VMs. The discount is applied to the vCPU and memory usage of those VMs, not to the node premium. So you can get SUDs on the compute resources used by VMs on sole-tenant nodes.
The discount is based on cumulative usage. If you stop a VM, the usage stops accumulating. The total usage for that resource type is the sum of all running hours across all VMs. If you stop a VM for part of the month, your total usage decreases, potentially lowering your discount tier. For example, if you run a VM for 400 hours, then stop it for 100 hours, then run it again for 300 hours, total usage is 700 hours. In a 730-hour month, that's 95.9% usage, which gives a discount of about 28.5% (linear interpolation between 20% at 75% and 30% at 100%). If you had run it continuously for 730 hours, you'd get 30%. So stopping reduces the discount slightly.
SUDs are applied automatically. There is no configuration required. When you create a VM, it is automatically eligible for SUDs if it meets the criteria (not preemptible, not spot, not covered by CUD). The discount is calculated at the end of the billing month and reflected in your invoice. You do not need to opt in or sign up.
You've just covered Sustained Use Discounts (SUDs) — now see how well it sticks with free ACE practice questions. Full explanations included, no account needed.
Done with this chapter?