GPU-accelerated instances deliver superior computing performance and graphics processing for demanding workloads.
Procedure
-
Go to the instance creation page.
-
Click the Custom Launch tab.
-
Configure the billing method, region, network, availability zone, instance type, and image.
Each parameter is described in the Parameters section.
-
Review the instance configuration and set options such as usage duration.
The usage duration settings vary based on the billing method of the instance:
-
For a pay-as-you-go or spot instance: Set an automatic release time, or manually release the instance or configure an automatic release schedule after creation. Release an instance.
-
For a subscription instance: Select a duration and specify whether to enable auto-renewal. You can also renew manually after creation. Renew a subscription instance.
-
-
Read the terms of service and agreements, and then click Confirm Order.
-
On the payment page, verify the total cost and complete the payment.
Parameters
Billing method
The billing method determines pricing and resource lifecycle rules.
To purchase pay-as-you-go ECS resources, your account balance (cash balance) plus coupon value must be at least CNY 100.00. Top up your account.
|
Billing method |
Description |
References |
|
Subscription |
You pay for resources before you use them. The minimum subscription duration is one week. |
|
|
Pay-as-you-go |
Billed per second. Purchase and release resources on demand. Note
Use a savings plan or a reserved instance to optimize costs. |
|
|
Spot Instance |
Discounted compared to pay-as-you-go. Billed after use, but may be reclaimed due to price changes or insufficient inventory. |
Region
A region is the geographical location of a data center. Select a region close to your users to reduce network latency. You cannot change the region after an instance is created. For more information, see Regions and zones.
Network and availability zone
We recommend using a Virtual Private Cloud (VPC). VPCs are logically isolated network environments that provide enhanced security and support features such as Elastic IP Addresses (EIPs), elastic network interfaces, and IPv6.
A zone is a physical area within a region that has independent power and networking. Instances in the same zone have lower network latency and provide faster access for users.
|
Network type |
Description |
Related topics |
|
VPC |
A VPC is an isolated network environment that you can define on Alibaba Cloud. You have full control over your VPC, such as specifying IP address ranges and configuring route tables and gateways. If you do not need to customize VPC settings during instance creation, you can skip this step. The system automatically creates a default VPC and vSwitch. Select an existing VPC and vSwitch, or click Create VPC or Create vSwitch to go to the VPC console and create them. After creating them, return to the ECS instance creation page and click the Note
If you need to assign an IPv6 address to the instance, select a VPC and vSwitch for which an IPv6 CIDR block is enabled. |
Instance type and image
The instance type and image define basic instance attributes: vCPUs, memory, and operating system.
Instance type
Available instance types vary by factors such as the region. Go to the Instance types available for purchase page to view the instance types that you can create.
If you have specific configuration requirements, such as attaching multiple elastic network interfaces, using ESSDs, or using local disks, make sure that the selected instance type supports these features. For more information about the features, use cases, and specifications of instance families, see Instance families.
If you set Billing Method to Spot Instance, configure the usage duration and maximum price.
-
Instance Usage Duration: specifies the protection period for a Spot Instance. After the protection period ends, the instance may be automatically released due to market price fluctuations or insufficient inventory of the instance type.
Duration
Description
1 Hour
The Spot Instance is protected for one hour after it is created and will not be automatically released during this period.
None
The Spot Instance has no protection period after creation but is more cost-effective than a Spot Instance with a protection period.
-
Highest Price per Instance:
Maximum price
Description
Use Automatic Bid
Always use the real-time market price for the instance type. The market price does not exceed the price of the corresponding pay-as-you-go instance. This option prevents the Spot Instance from being released because the market price exceeds your bid, but it does not prevent release due to insufficient inventory of the instance type.
Set Maximum Price
Enter a specific price cap. The Spot Instance is automatically released if the real-time market price exceeds your cap or if inventory becomes insufficient.
After you configure the instance, you can confirm the details of the selected instance type under Selected.If you want to try FastGPU for one-click AI task deployment on Alibaba Cloud IaaS, click Learn More >. What is FastGPU?.
Image
An image provides the data required to run an instance. Alibaba Cloud offers images from the following sources.
|
Image source |
Description |
|
Licensed base images provided by Alibaba Cloud, including Windows Server and mainstream Linux distributions. |
|
|
An image you create or import, containing your system and application configurations for reuse. |
|
|
A custom image shared by another Alibaba Cloud account for cross-account instance creation. |
|
|
Pre-reviewed images from Alibaba Cloud Marketplace for one-click deployment of websites, applications, and other environments. |
|
|
A fully public image published by an Alibaba Cloud user. You can use community images from others or publish your own. |
You can auto-install an NVIDIA Tesla driver during image selection, or use a free image with a pre-installed NVIDIA Tesla driver or NVIDIA GRID driver. Automatically install or load a Tesla driver when you create a GPU-accelerated instance. Automatically load a GRID driver from a community image that has a pre-installed driver.
Storage
Instances gain storage by adding system disks, data disks, elastic ephemeral disks, and Apsara File Storage NAS (NAS) file systems. ECS offers both cloud disks and local disks to meet the requirements of different scenarios.
-
Cloud disks can be used as system disks and data disks. Types include ESSDs, standard SSDs, and ultra disks. For more information, see Disks overview.
NoteA cloud disk created with an instance has the same billing method as the instance.
-
Local disks can be used only as data disks. If the instance type is equipped with local disks, such as local SSD or big data instance families, the local disk information is displayed on the page. For more information, see Local disks.
NoteYou cannot manually attach local disks to an instance.
System disk
The system disk is used to install the operating system. The default capacity is 40 GiB, but the minimum configurable capacity depends on the image type, as shown in the following table.
|
Image |
Capacity range (GiB) |
|
Linux (excluding FreeBSD and Red Hat) |
[max{20, image size}, 2048] |
|
FreeBSD |
[max{30, image size}, 2048] |
|
Red Hat |
[max{40, image size}, 2048] |
|
Windows |
[max{40, image size}, 2048] |
(Optional) Data disk
Data disks are used to store application data. When you select a data disk, you can also encrypt the cloud disk to meet data security or regulatory compliance requirements. For more information about data encryption, see Encryption.
The number of data disks that can be attached to a single instance is limited. For more information, see Block storage limits.
(Optional) Snapshots
A snapshot is a backup of a cloud disk's data at a specific point in time. Creating a cloud disk from a snapshot is an efficient way to import data. You can enable automatic backups for cloud disks during instance creation to protect against data loss from accidental deletion.
Select an existing automatic snapshot policy, or click Create Automatic Snapshot Policy to go to the Snapshot page and create one. For more information, see Create an automatic snapshot policy. After the policy is created, return to the ECS instance creation page and click the 
icon to refresh the automatic snapshot policy list.
Snapshots incur fees. For more information, see Snapshot billing.
(Optional) NAS
If you need to share a large amount of data among multiple instances, we recommend that you use a NAS file system to reduce data copying and synchronization costs.
Select an existing NAS file system, or click Create File System to go to the NAS console and create one. For more information, see Create a General-purpose NAS file system in the console. After the file system is created, return to the ECS instance creation page and click the icon to refresh the NAS file system list. For information about how to mount a NAS file system, see Mount a NAS file system when you purchase an ECS instance.
Network and security group
Configure public internet access, inter-resource communication, and network security.
(Conditionally required) Public IP address
An instance must have a public IP address to communicate with the internet. You can automatically assign a fixed public IP address when you create the instance, or configure one later using an EIP or NAT gateway. EIPs and NAT gateways must be purchased separately. For more information, see What is an Elastic IP Address? and What is a NAT Gateway?.
Select Assign Public IPv4 Address, and set Bandwidth Billing Method and Bandwidth or Maximum Bandwidth.
For detailed rules on public bandwidth billing, see Public bandwidth billing.
|
Billing method |
Description |
|
Pay-by-bandwidth |
Charges are based on a fixed bandwidth that you specify. The actual outbound bandwidth does not exceed this value.
|
|
Pay-by-traffic |
Charges are based on the actual data transferred. To avoid high traffic fees, you can set a peak outbound bandwidth.
(Optional): Select Upgrade to CDT for Data Transfer Billing. Cloud Data Transfer (CDT) provides an efficient and economical solution for managing public bandwidth costs with flexible billing, free data transfer quotas, tiered pricing, and unified billing across multiple products. It offers discounts compared with the standard pay-as-you-go method. For more information, see What is Cloud Data Transfer (CDT)? Important
|
Security group
A security group acts as a virtual firewall that controls inbound and outbound traffic for instances within the group. For more information, see Security groups overview.
If no security group exists in the selected VPC, the system creates a default security group. By default, the security group allows inbound traffic on ports 22 and 3389, and via the ICMP protocol. You can also allow traffic on ports 80 and 443 as needed, or modify the security group rules after creation.
You can also select an existing security group or create a Create Security Group. When you create a new security group, you must configure the Security Group Name, Security Group Type, and Open IPv4 Ports/Protocols.
For a detailed description of security group settings, see Create a security group.
(Optional) Elastic network interface
An elastic network interface can be a primary ENI or a secondary ENI. The primary ENI cannot be unbound from the instance and is created and released with the instance. Secondary ENIs can be freely bound to and unbound from instances, allowing you to shift network traffic between them. To create a secondary ENI along with the instance, click the 
icon and select the vSwitch for the secondary ENI.
You can add only one secondary ENI during instance creation. You can create and bind additional secondary ENIs after the instance is created. For the number of ENIs supported by each instance type, see Instance families.
(Optional) Configure IPv6
IPv6 provides a vast number of addresses, solving IP address exhaustion and removing connection barriers for various devices.
Select Assign IPv6 Address Free of Charge. After an IPv6 address is assigned, you must log on to the instance and configure the IPv6 address in the operating system to use it. For instructions, see IPv6 communication.
Management settings
Configure logon credentials for remote access and tags for resource management.
Logon credential
Logon Credential are used to securely log on to an instance. For more information about instance connection methods, see Connect to an ECS instance.
|
Method |
Description |
|
Key Pair Note
Only Linux instances support key pair authentication. |
Select a username for logon and an existing key pair, or click Create Key Pair to create one. After creating the key pair, return to the ECS instance creation page and click the You can set the username to root or ecs-user. Warning
The root user has the highest operating system privileges. Using root as the username can pose security risks. We recommend that you use the standard user ecs-user. |
|
Use Predefined Password Note
This authentication method is supported only for Custom Image and Shared Image. |
You can use the preset password of the selected image for logon. Make sure that a password is set in the selected image. |
|
Password |
Enter and confirm a password. The username for password-based logon is determined as follows:
|
|
Set Later |
After the instance is created, you can bind a key pair or reset the instance password. For instructions, see Bind an SSH key pair to an instance and Reset the logon password of an instance. |
Tag
A Tag is a key-value pair for identifying the created instances, cloud disks, and primary ENIs. Tags make resources easier to search for and manage. You can select an existing tag or enter a tag key and value to create one. For more information about tags, see Tags.
(Optional) Advanced options
Customize instance hostname, metadata, and user data.
|
Parameter |
Description |
|
Instance Name, Description, Host Name, and Sequential Suffix |
Set sequential names and hostnames when creating multiple instances to identify batches. Batch configure sequential instance names or hostnames. |
|
Instance RAM Role |
Attach an Instance RAM Role to grant the instance temporary STS tokens for accessing cloud service APIs and resources, improving security. Select an existing Instance RAM Role, or click Create Instance RAM Role to create one in the RAM console. After creation, click the |
|
Metadata access mode |
Instance metadata provides information about the instance in the Alibaba Cloud system. Query it from within the instance to configure or manage it. Instance metadata. |
|
User data |
Provide a startup script to automate instance configuration, or pass plain data to the instance. Customize instance bootstrap configurations. If you did not select the Images option in the Image section and you are familiar with automatic driver installation scripts, you can enter an automatic installation script in the User data section to install the NVIDIA Tesla driver. For more information, see Install a driver by using an automatic installation script. Note
Enter your prepared instance user data in the input box. If the user data is Base64-encoded, select Enter Based64-encoded Content. |
|
Resource Group |
Organize resources across regions and services by business unit, with group-level permission control. Resource groups. Select an existing resource group, or click Create Resource Group to create one. After creation, click the |
|
Deployment Sets |
Distribute instances across physical servers for high availability and disaster recovery. Select an existing deployment set, or click Manage Deployment Sets to create one. After creation, click the |
|
Dedicated Hosts |
A Dedicated Host (DDH) is a single-tenant physical server for strict security, compliance, and Bring Your Own License (BYOL) needs. Select an existing DDH, or click Create Dedicated Host to create one. After creation, click the |
|
Private Pool Type |
Elastic assurances and capacity reservations generate private pools that guarantee resource availability for specific instance types. Resource Assurance and Reservation overview. Note
Elastic assurance and capacity reservation guarantee resource supply certainty only for pay-as-you-go instances.
|
Next steps
-
Connect to the instance
Connect using Workbench, VNC, or third-party tools. Choose an ECS remote connection method.
-
Install a driver
If no driver was auto-installed during instance creation, install the correct Tesla or GRID driver for your instance type. Installation guidelines for Tesla and GRID drivers.
Related documents
-
Create instances programmatically: RunInstances or CreateInstance.
-
Enable, hibernate, restart, release, or stop an instance: Manage a GPU-accelerated instance.
-
Deploy an NGC environment for immediate access to optimized deep learning frameworks. Deploy an NGC environment to build a deep learning development environment.
-
Troubleshoot GPU issues: FAQ about GPU-accelerated instances.