Elastic GPU Service

Powerful parallel computing capabilities based on GPU technology.

Buy Now Contact Sales


Elastic GPU Service (EGS) is a powerful GPU-based computing service suited for high-demand computing, including video decoding, graphics rendering, deep learning, and scientific computation. The service boosts fast speed, strong concurrent computing, and floating-point computing capabilities.


Graphic Design

Ideal for graphic engineering design, non-linear editing, distance learning applications, and 3D presentations.

Video Rendering

Conduct large-scale HD video transcoding, 4K live broadcasting, multi-party video conferencing, and video signal processing.


Excellent for animation rendering, digital image processing, financial computing applications, and scientific computing.

Deep Learning

Highly suitable for image recognition, speech recognition, video content identification, source film repair, and other applications.

Product Details

Elastic GPU Service is an online computing service that provides reliable, scalable and on-demand distributed computing capacity. Accelerated by GPU on the cloud, EGS allows you to harness powerful parallel computing capabilities and the elasticity to automate and adapt to an ever-changing business environment in real-time.

Meet Network Performance Requirements

GPU instances support a maximum of 650,000 PPS and 6 Gbps intranet bandwidth, meeting high network performance requirements.

Strong Rendering Performance

Use Ultra Cloud Disk or SSD Cloud Disk to maximize computing and rendering performance, while ensuring high data availability.

GPU Instance


GN5 is built on the Nvidia Tesla P100 GPU. The P100 GPU provides up to 74.4 TFLOP/s of computing power for single-precision floating-point operations, enough to meet the demands for deep learning and massively parallel floating-point operations.

P100 GPU also provides up to 37.6 TFLOP/s of computing power in double-precision floating-point operations, fulfilling requirements for scientific and high-performance scenarios. Supported by GPU Direct P2P technology, the PCI bus can be utilized to implement direct communication between GPUs, greatly reducing latency.

Lastly, GN5 integrates perfectly with other elastic computing products to power common solutions. It is also suited for containerized service usage, to simplify the deployment, operation, and maintenance of resource scheduling services.

gn5, compute optimized type family with GPU


NVIDIA P100 GPU processors

No fixed ratio of vCPU to memory

High-performance ephemeral NVMe SSD Disk

2.5 GHz Intel Xeon E5-2682 v4 (Broadwell) processors

Matching network performance to instance computing type (the more advanced the computing type, the more powerful the network performance),

Ideal for:

  • Deep learning

  • Scientific computing, such as computational fluid dynamics, computational finance, genomics, and environmental analysis

  • High-performance computing, rendering, multi-media coding and decoding, and other server-side GPU compute workloads

Instance types

Instance TypevCPUMemory (GiB)Ephemeral Storage (GiB)GPUIntranet Bandwidth (Gbit/s)Packet Forwarding Rate (10 thousand PPS)*NIC Queues
ecs.gn5-c4g1.xlarge4304401 * NVIDIA P1003301
ecs.gn5-c8g1.2xlarge8604401 * NVIDIA P1003301
ecs.gn5-c4g1.2xlarge8608802 * NVIDIA P1005100***2
ecs.gn5-c8g1.4xlarge161208802 * NVIDIA P1005100***2
ecs.gn5-c28g1.7xlarge281124401 * NVIDIA P1005100**8
ecs.gn5-c8g1.8xlarge3224017604 * NVIDIA P10010200**8
ecs.gn5-c28g1.14xlarge562248802 * NVIDIA P10010200**14
ecs.gn5-c8g1.14xlarge5448035208 * NVIDIA P10025400**14

* For more information about PPS testing, please see Test network performance.

** Testing conditions: For Centos 7.3, no more than vCPU core number / 4 queues for NICs should be enabled. If you wish to adjust multiple queues for NIC, please restart the instance.

*** Testing conditions: Enable two queues for NICs with CentOS 7.3. If you want to adjust multiple-queue for NIC, please restart the instance.

Please note that you cannot change gn5 instance configurations.


Elastic GPU Service offers different payment options to suit your individual needs, including Pay-As-You-Go and Subscription. The following prices are for reference only. The exact price is based on the customized plan you select.


A. Annual payment

Choose to pay yearly to enjoy the biggest discount.

B. Monthly payment

Pay per month for instances, which lowers one-time investment for computing resources and offers a relatively low price per unit hour.

C. Hourly payment

Pay per hour for instances to meet short-term and temporary requirements at the lowest single time cost.

Getting Started

Create a gn5 Instance

Create gn5 instance as explained by create an instance in the Quick Start guide for Elastic Compute Service (ECS). When creating the instance, please note the following items:

Region: Currently, gn5 is only available in the following regions: China East 1, China East 2, China North 2, China South 1, US East 1 (Virginia), US West 1, Hong Kong, Asia Pacific SE 1, Asia Pacific SE 2, and Germany 1.

Network Type: Select VPC, gn5 is available for Virtual Private Cloud network.

Instance Type: Select GPU Compute Type gn5) under Generation III.

Network Bandwidth: Select the peak bandwidth as required.
If the Windows 2008 R2 image is in use and you wish to connect to a gn5 instance, you must enable Internet access for the instance as you cannot connect to a gn5 instance through the Management Terminal of the ECS management console. If you wish to allocate an Internet IP address to the instance, do not set the peak bandwidth to 0 Mbps.

Image: Select the image as required.

GPU Driver Download and Installation

Before using a gn5 instance, you first need to install a GPU driver. To download and install the GPU driver please review the following steps.

1. Navigate to NVIDIA official website to download the corresponding driver for the operating system and P100 GPU. The download URL is: http://www.nvidia.com/Download/index.aspx?lang=en-us.

2. Manually locate the driver for the instance. Set the parameters as follows:

o Product Type: Tesla

o Product Series: P-Series

o Product: Tesla P100

o Operating System: The corresponding version according to the instance image.
If the operating system is not displayed in the drop-down list, click Show all Operating Systems at the bottom of the drop-down list.
If the instance uses a Linux image that is not in the list, select Linux 64-bit.

3. Click SEARCH.

4. After confirming the information, click DOWNLOAD.

5. Install the GPU driver by following the ADDITIONAL INFORMATION on the download page.
Please see Linux 64-bit OpenSUSE 13.2 as an example:


For Windows 2008 R2 or earlier version, if you enter the Management Terminal by clicking Connect in the ECS console after the GPU driver is installed, the Management Terminal will be stuck at either a black screen or the startup interface. If the instance can access the Internet, you need to connect to the ECS instance remotely using other protocols, such as the Remote Desktop Protocol developed by Microsoft.


Elastic GPU Service is an online computing service that provides reliable, scalable and on-demand distributed compute capacity on the cloud. GPU allows you to have complete control of your computing environment and helps to automate and meet your business needs in real time.

Developer Resources

Developer Tools

GPU Forum