Getting Started

Last updated:2026-01-16 16:58:48
Getting Started
Last updated: 2026-01-16 16:58:48
Purchase NVIDIA GPU Instance
Purchase Note
Before purchasing, please ensure you are familiar with the Tencent Cloud GPU Service and understand configuration and pricing, and purchase based on actual needs.
Ensure you understand the AZ of the selected GPU instance. For AZ information, see GPU Instance Types.
For more information on configuration options during the purchase process, please refer to Purchasing a Customized Linux Instance.
Purchase Procedure
This guide uses the GN10Xp instance type as an example to guide how to quickly purchase a GPU instance:
Step 1: Log In to the Purchase Page
﻿
Step 2: Select Billing Mode, Network, Region, and Instance Type
After going to the purchase page, select the billing mode, region (AZ), and instance. As shown in the following figure:
﻿
Billing Mode: yearly/monthly subscription, spot instances, pay-as-you-go or underwriting. For details, see Billing Overview.
Region and AZ: You can select the AZ based on the actual purchase page. For details, see Instance Types.
Architecture: Select Heterogeneous Computing.
Type: This guide uses GPU computing instances GN10Xp as an example. You can refer to Instance Types to select the most suitable instance configuration.
Step 3: Select Images and Storage
1. Select the image for the GPU.
GPU currently supports three image types: Public Images, Custom Images, and Shared Images. For details, see Image Overview.
New Tencent Cloud users can select Public Images and choose the version as needed. As shown in the following figure:
﻿
Note:
GPU must have the corresponding GPU driver installed to function properly. You can install the driver using the following methods:
If you select Public Images, select Install GPU drive automaticlly to pre-install the corresponding driver version. We recommend using this method. It is only supported for some Linux public images. For details, see NVIDIA Driver Installation Guide.
If you select Public Images, you can manually install the corresponding driver by referring to NVIDIA Driver Installation Guide after the GPU instance is successfully created.
If you purchase vGPU instances, you could select public images with pre-installed GRID drivers, with no need to install drivers separately.
2. Select the storage for the Cloud GPU Service. Both the system disk and data disk types and sizes can be flexibly selected. As shown in the following figure:
﻿
3. Main Type Information is as follows:
Balanced SSD, Enhanced SSD, CloudSSD, Premium Cloud Disk. For CBS types and corresponding billing rules, see CBS Price Overview.
4. After the setting is completed, click Next: Configure Network and Host.
Step 4: Configure Network, Security Group, and Host
1. Select the network and bandwidth for the Cloud GPU Service. As shown in the following figure:
﻿
Network: Select an existing virtual private cloud (VPC) or create a new one.
Public IP: If your instance requires public network access capabilities, you need to select this option, and a public IP address will be assigned to the instance after creation.
Line type: All public IPs are BGP IPs, which can fully utilize the advantages of Tencent Cloud's DDoS network: low latency and high protection.
Bandwidth billing mode: For public network bandwidth, the following three billing modes are available:
Monthly-subscribed bandwidth: Select a fixed bandwidth. When the bandwidth is exceeded, packets will be dropped. Suitable for scenarios with low network fluctuations.
Bill by traffic: Billed based on actual traffic usage. You can limit the peak bandwidth. Packets will be dropped when the instantaneous bandwidth exceeds the limit (suitable for scenarios with significant network fluctuations).
Bandwidth package: The bandwidth upper limit of the IP address is affected by the Bandwidth package. The actual limit is the lower of the two. For details, see Bandwidth package Billing Overview.
Bandwidth cap: The upper limit of the instance's public network bandwidth, set it as needed.
2. Create a new security group or select an existing one to control the access scope of ports. As shown in the following figure:
﻿
3. Set the login password for the Cloud GPU Service.
Set password: Password login is set. You can log in to the CVM console to reset the password.
SSH key pair: Tencent Cloud recommends that you use an SSH key pair to log in to Linux instances. View the SSH Keys Guide.
Reset password after creation: After the instance is created, log in to the CVM console and "reset the password" to complete the setup.
4. Click Next: Confirm Configuration.
Step 5: Confirm Configuration
1. Please verify the following content in the Confirm Configuration step. As shown in the following figure:
﻿
Verify that the configurations, including instance type, image, storage, bandwidth, and security group, meet your requirements.
Please select and verify the purchase quantity and duration.
2. Read and select Agree to the "Tencent Cloud Service Terms" and Purchase Notes, then click Buy Now.
Step 6: Verify the Order and Payment
Check the order information and select a payment method to pay.
After the payment is successful, go to the Console. You can log in to the instance once it has been created and started.
Getting Started with a Windows GPU Instance
The steps for creating and configuring a Windows GPU instance are the same as those for a Windows CVM instance. For details, see Purchasing a Customized Windows Instance.
Getting Started with a Linux GPU Instance
The steps for creating and configuring a Linux GPU instance are the same as those for a Linux CVM instance. For details, see Purchasing a Customized Linux Instance.
Appendix
GPU Driver Pre-installation Instructions
Cloud GPU Service support automatic installation of GPU drivers for specific Linux public images. Upon initial startup, the GPU driver, CUDA, and cuDNN libraries will be automatically installed based on the selected version. This process takes approximately 15 to 20 minutes. Alternatively, you can manually install the required driver version. For details, please refer to Driver Version Selection.
Using API to Create a GPU Instance
If you need to create a GPU instance via API, the server provides the interface (RunInstances). Request domain: cvm.intl.tencentcloudapi.com. For details, see RunInstances.