A resident resource pool reserves GPU computing capacity through a monthly subscription. Purchase a pool, assign resident instances to your GPU functions, and get predictable costs with zero cold starts.
How it works
Resident resource pools are exclusive to GPU functions and use subscription billing. By reserving GPU cards in advance, you lock in scarce GPU capacity and keep your services running without resource shortages.
After you purchase a pool and attach resident instances to a function, that function processes all requests through its resident instances. The maximum number of concurrent requests is:
(Number of allocated resident instances) x (Instance concurrency)
Requests within this limit receive a real-time response with no cold start latency. Requests that exceed this limit are throttled.
Functions that use resident instances cannot use on-demand instances at the same time.
Use cases
Resident resource pools fit workloads with the following characteristics:
Sustained GPU demand: Continuous, stable resource usage where reserved capacity is cost-effective.
Latency-sensitive inference: Zero cold starts for real-time response times.
Predictable budgets: Fixed monthly costs with no variable billing surprises.
Billing
Resident resource pools use a monthly subscription model. Pay upfront for one month, several months, or a year. When a function uses resident instances from a pool, no additional on-demand fees apply beyond the subscription cost.
For current prices, see the purchase page.
Billable items
Each resident resource pool incurs two types of fees:
| Fee type | Description |
|---|---|
| GPU card fees | Covers GPU memory, vCPUs, and memory. Only full cards can be purchased. See GPU full card specifications and pricing for details. |
| Disk fees | Additional disk capacity per card. Range: 10 GB to 60 GB in 10 GB increments. Pricing varies by region. |
Disk pricing by region:
| Region | Disk price |
|---|---|
| China (Hangzhou), China (Shanghai), China (Beijing), China (Ulanqab), China (Shenzhen) | USD 0.17/GB/month |
| US (Virginia), Germany (Frankfurt), Singapore | USD 0.21/GB/month |
GPU full card specifications and pricing
After you purchase a resident resource pool, Function Compute converts your total purchased capacity into an available pool that you can flexibly allocate across functions as resident instances.
China regions (Hangzhou, Shanghai, Beijing, Ulanqab, Shenzhen)
| GPU card type | GPU memory | vCPU | Memory | Disk | Monthly price |
|---|---|---|---|---|---|
| Ada series | 48 GB | 8 | 64 GB | 10 GB | USD 1,026 |
| Ada.2 series | 24 GB | 8 | 32 GB | 10 GB | USD 908 |
| Ada.3 series | 48 GB | 8 | 64 GB | 10 GB | USD 1,320 |
| Xpu.1 series | 96 GB | 8 | 96 GB | 10 GB | USD 1,507 |
| Hopper series | 96 GB | 16 | 96 GB | 10 GB | USD 2,887 |
US (Virginia)
| GPU card type | GPU memory | vCPU | Memory | Disk | Monthly price |
|---|---|---|---|---|---|
| Ada series | 48 GB | 8 | 64 GB | 10 GB | USD 1,048 |
| Hopper series | 96 GB | 16 | 96 GB | 10 GB | USD 3,027 |
Germany (Frankfurt)
| GPU card type | GPU memory | vCPU | Memory | Disk | Monthly price |
|---|---|---|---|---|---|
| Ada series | 48 GB | 8 | 64 GB | 10 GB | USD 1,576 |
| Hopper series | 96 GB | 16 | 96 GB | 10 GB | USD 4,032 |
Singapore
| GPU card type | GPU memory | vCPU | Memory | Disk | Monthly price |
|---|---|---|---|---|---|
| Ada series | 48 GB | 8 | 64 GB | 10 GB | USD 1,289 |
| Hopper series | 96 GB | 16 | 96 GB | 10 GB | USD 3,537 |
For current prices, see the purchase page.
Manage resident resource pools
Purchase a resident resource pool
Log on to the Function Compute console. In the left-side navigation pane, choose Elastic Management > Resident Resource Pools. In the top menu bar, select a region.
On the Resident Resource Pools page, click Purchase Resident Resource Pool. On the purchase page, configure the following parameters:
Parameter Description Region The region where the pool is deployed. GPU Card Type The GPU card series to use. Number of Cards The number of GPU cards in the pool. Disk Specification per Card The disk capacity allocated per card (10 GB to 60 GB, in 10 GB increments). Subscription Duration The billing period: 1, 2, 3, or 6 months, or 1 year. Click Buy Now.
View resident resource pools
On the Resident Resource Pools page, view and filter all purchased pools. Each entry shows function allocation details, remaining quota, and expiration time so you can allocate resources efficiently and avoid waste.
Scale out (upgrade) a resident resource pool
On the Resident Resource Pools page, find the target pool and click Scale-out in its row.
Follow the on-screen instructions to complete the upgrade.
Only the number of cards and the disk specification per card can be increased. The GPU card type cannot be changed, and neither the number of cards nor the disk specification can be decreased. Scale out with caution to avoid resource waste or incorrect configuration. Expired pools must be renewed and activated before scaling out.
Expiration and renewal
Renew your resident resource pool before it expires or its resources are depleted.
Expiration reminders
Message Center sends renewal reminders 7, 3, and 1 day before your resident resource pool expires.
What happens when a pool expires
Automatic shutdown: The pool shuts down immediately and all resident instances for its associated functions are released.
Function impact: If you do not switch the function's instance type to on-demand, the function becomes unavailable.
90-day grace period: After 90 days in a shut-down state, the pool is permanently released and automatically detached from all associated functions. To resume, purchase a new pool.
Message Center sends release reminders 30, 7, and 1 day before the pool is permanently released.
Renewal options
Auto-renewal
Enable auto-renewal in either of the following ways:
During purchase, select the Enable Auto-renewal on Expiration checkbox.
After purchase, log on to the Renew page, find the target resource, and enable or disable auto-renewal.
Manual renewal
On the Resident Resource Pools page, find the target pool and click Renew in its row. Follow the on-screen instructions to complete the renewal.
View bills
After a bill is generated, view consumption details in Expenses and Costs - Billing Details. For more information, see Query bills.
Unsubscription
To terminate a resident resource pool, Function Compute revokes the resources and issues a refund based on Alibaba Cloud unsubscription rules. Before you request an unsubscription, review the effects and calculation rules below.
Before you unsubscribe
Detach the pool from all functions first. Log on to the Function Compute console and set the number of resident instances to 0 for every function attached to the target pool.
Effects of unsubscription
An unsubscription cannot be revoked once successful. Verify the order information carefully.
If the order was part of a promotion, you cannot participate in the same promotion again after unsubscribing.
Vouchers and coupons used in the original order are voided.
Refund calculation
Refund amount = Actual amount paid - Consumed amount
Actual amount paid: The cash amount paid for the order. This excludes any value from vouchers or coupons.
Consumed amount: Calculated based on your usage duration:
| Usage duration | Consumed amount formula |
|---|---|
| Less than 30 days | (Daily price x Usage days) x 1.2 |
| 30 days or more | Daily price x Usage days |
If a renewal order has not yet taken effect, you can only unsubscribe from the resource itself. You cannot separately unsubscribe from the pending renewal order.
Request an unsubscription
Log on to the Unsubscribe page.
Find the target resource and click Unsubscribe in the Actions column.
Follow the on-screen instructions to complete the unsubscription.
Make sure the number of resident instances is set to 0 for all attached functions before submitting the request. For more information, see Unsubscribe from an effective order.
Refund processing
The refund is typically processed within 2 to 3 business days. Only amounts paid in cash or with a stored value card are refunded. Portions paid with vouchers or coupons are not refunded. Track your refund status.
Overdue payments
If your account has overdue payments, existing purchased pools continue to operate normally. However, you cannot purchase new pools, upgrade existing pools, or renew them until the balance is settled. For more information, see Overdue payments.
FAQ
How do I use a resident resource pool?
After purchasing a pool, set the instance type to Resident Instance when creating a GPU function. This attaches the function to the pool and allocates resident instances. For step-by-step instructions, see Configure resident instances.
Can I scale in (decrease) a resident resource pool?
To avoid affecting online services, scaling in a resident resource pool is not currently supported. The number of GPU cards and the disk specification per card can only be increased, not decreased.
How do I set the right disk specification per card?
The disk specification per card is not directly tied to the disk specification in your function configuration. Function Compute converts the total disk you purchase into a shared pool of disk capacity that can be allocated across functions.
Example: Function 1 requires a 30 GB disk with 4 instances. Function 2 requires a 60 GB disk with 2 instances. The total disk needed is 30 GB x 4 + 60 GB x 2 = 240 GB. If you purchase 6 cards, set the disk specification per card to 240 GB / 6 = 40 GB.