Elastic Cloud Server

GPU Cloud Server

2026-03-23 09:40:23

GPU Type

GPU cloud servers can provide excellent floating-point computing capabilities, easily handling massive computational scenarios with high real-time and high concurrency requirements. Especially suitable for applications such as deep learning, scientific computing, CAE, 3D animation rendering, and CAD.

Currently, there are two major categories provided: Compute Acceleration and Graphics Acceleration.

Among them, the compute acceleration type of GPU cloud server currently supports both Nvidia and Ascend graphics cards; the graphics acceleration foundation type supports Nvidia graphics cards.

GPU Computing Accelerated

The GPU computing acceleration type of GPU cloud server utilizes GPU hardware pass-through technology, mainly suitable for scenarios such as AI deep learning training, inference, scientific computing, video transcoding, and graphic processing.

On Sale: PI7 and PI2

Features of Compute Acceleration GPU Cloud Servers

Specification Name

GPU Model

Quantity of GPUs

Single GPU Performance

Disk Type

P8A

Nvidia   Tesla A100 (40G PCIE)

1,2,4

19.5   TFLOPS Single Precision Floating Point Calculation
  9.7 TFLOPS Double Precision Floating Point Calculation
  156 TFLOPS TF32 AI Acceleration

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

PI7

Nvidia   Tesla A10

1,2,4

31.2   TFLOPS Single Precision Floating Point Calculation
  62.5 TFLOPS TF32AI Acceleration

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

P2V

Nvidia   Tesla V100

1,2,4

14   TFLOPS Single Precision Floating Point Calculation
  7 TFLOPS Double Precision Floating Point Calculation
  112 TFLOPS TF32 AI Acceleration

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

P2Vs

Nvidia   Tesla V100s

1,2,4

16.4   TFLOPS Single Precision Floating Point Calculation
  8.2 TFLOPS Double Precision Floating Point Calculation
  130 TFLOPS Double Precision Floating Point Calculation

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

PI2

Nvidia   Tesla T4

1,2,4

8.1   TFLOPS Single Precision Floating Point Calculation
  130 TOPS INT8 Calculation
  260 TOPS INT8 Calculation

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

PAK1

Atlas   300I pro

1,2,4

70   TFLOPS Half Precision Floating Point Calculation
  140 TOPS INT8 Computation

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

P8A Type Cloud Server

The P8A model cloud server is equipped with an NVIDIA A100 40GB PCIe GPU and utilizes GPU pass-through technology. While providing the flexibility of a cloud server, it offers high-performance computing capabilities and excellent cost-effectiveness. A single card can provide up to 77.97 TFLOPS of half-precision floating-point computing power and 9.746 TFLOPS of double-precision floating-point computing power. The P8A model cloud server can provide ultra-high general computing capabilities, suitable for AI deep learning and scientific computing. It can demonstrate significant computational advantages in fields such as deep learning training, scientific computing, computational fluid dynamics, computational finance, seismic analysis, molecular modeling, and genomics.

Specification    Name

vCPU

Memory    (GB)

GPU

Graphics Memory (GB)

Virtualization    Type

Maximum Bandwidth (Gbps) / Baseline Bandwidth (Gbps)

Number of NIC Queues

Maximum Packet Handling Capacity (in 10,000 PPS)

p8a.6xlarge.4

24

96

1×A100

1×40GB

KVM

30/11

8

300

p8a.12xlarge.4

48

192

2×A100

2×40GB

KVM

36/23

16

600

p8a.24xlarge.4

96

384

4×A100

4×40GB

KVM

47/45

32

1000

Common Software Compatibility List

The P8A model cloud server is primarily designed for compute-acceleration scenarios, such as deep learning training, inference, scientific computing, molecular modeling, seismic analysis, and other similar applications. If the application software utilizes the CUDA parallel computing capabilities of the GPU, the P8A model cloud server can be used. The common software support list is as follows:

  • Common deep learning frameworks:      TensorFlow, Caffe, PyTorch, MXNet

  • RedShift for Autodesk 3dsMax, V-Ray for      3ds Max, and other CUDA-enabled GPU rendering software

Tips for Use

The P8A model cloud server currently supports the following versions of operating systems:

  • Windows Server 2019 Datacenter 64bit

  • Windows Server 2016 Datacenter 64bit

  • CentOS 8.1 64bit

  • CentOS 7.8 64bit

  • Ubuntu Server 20.04 64bit

  • Ubuntu Server 18.04 64bit

PI7 model cloud server

The PI7 model cloud server is equipped with the NVIDIA A10 Tensor Core GPU, which is specifically designed for AI inference. With GPU pass-through technology, it can provide extremely powerful real-time inference capabilities. The PI7 model elastic cloud server, with the help of the A10, can provide a maximum of 31.2 TFLOPS of FP32 computing power per card.

Specification    Name

vCPU

Memory    (GB)

GPU

Graphics Memory (GB)

Virtualization    Type

Maximum Bandwidth (Gbps) / Baseline Bandwidth (Gbps)

Number of NIC Queues

Maximum Packet Handling Capacity (in 10,000 PPS)

pi7.4xlarge.4

16

64

1×A10

1×24GB

KVM

17/7.5

8

200

pi7.8xlarge.4

32

128

2×A10

2×24GB

KVM

25/15

16

400

pi7.16xlarge.4

64

256

4×A10

4×24GB

KVM

47/45

32

800

Commonly Supported Software List

The PI7 model is primarily used for GPU inference computing scenarios, such as image recognition, speech recognition, natural language processing, and other similar applications. It can also support lightweight training scenarios and video encoding/decoding scenarios.

The common software support list is as follows:

  • Common deep learning frameworks:      TensorFlow, Caffe, PyTorch, MXNet

  • RedShift for Autodesk 3dsMax, V-Ray for      3ds Max, and other CUDA-enabled GPU rendering software

Tips for Use

The PI7 model cloud server currently supports the following versions of operating systems:

  • Windows Server 2019 Datacenter 64bit

  • Windows Server 2016 Datacenter 64bit

  • CentOS 8.1 64bit

  • CentOS 7.8 64bit

  • Ubuntu Server 20.04 64bit

  • Ubuntu Server 18.04 64bit

P2V model cloud server

The P2V model cloud server is equipped with the NVIDIA Tesla V100 PCIe GPU and utilizes GPU pass-through technology. While offering the flexibility of a cloud server, it delivers high-performance computing capabilities and excellent cost-effectiveness. The P2V model cloud server can provide ultra-high general computing capabilities, suitable for AI deep learning and scientific computing. It can demonstrate significant computational advantages in fields such as deep learning training, scientific computing, computational fluid dynamics, computational finance, seismic analysis, molecular modeling, and genomics.

Specification    Name

vCPU

Memory    (GB)

GPU

Graphics Memory (GB)

Virtualization Type

p2v.4xlarge.8

16

128

1*V100

1*32GB

KVM

p2v.8xlarge.8

32

256

2*V100

2*32GB

KVM

p2v.2xlarge.4

8

32

1*V100

1*32GB

KVM

p2v.4xlarge.4

16

64

2*V100

2*32GB

KVM

p2v.8xlarge.4

32

128

4*V100

4*32GB

KVM

Common Software Compatibility List

The P2V model cloud server is primarily designed for compute-acceleration scenarios, such as deep learning training, inference, scientific computing, molecular modeling, seismic analysis, and other similar applications. If the application software utilizes the CUDA parallel computing capabilities of the GPU, the P2V model cloud server can be used. The common software support list is as follows:

  • Common deep learning frameworks:      TensorFlow, Caffe, PyTorch, MXNet

  • RedShift for Autodesk 3dsMax, V-Ray for      3ds Max, and other CUDA-enabled GPU rendering software

Tips for Use

The P2V model cloud server currently supports the following versions of operating systems:

  • Windows Server 2019 Datacenter 64bit

  • Windows Server 2016 Datacenter 64bit

  • CentOS 8.1 64bit

  • CentOS 7.8 64bit

  • Ubuntu Server 20.04 64bit

  • Ubuntu Server 18.04 64bit

P2Vs model cloud server

The P2Vs model cloud server is equipped with the NVIDIA Tesla V100s PCIe GPU and utilizes GPU pass-through technology. While offering the flexibility of a cloud server, it delivers high-performance computing capabilities and excellent cost-effectiveness. The P2Vs model cloud server can provide ultra-high general computing capabilities, suitable for AI deep learning and scientific computing. It can demonstrate significant computational advantages in fields such as deep learning training, scientific computing, computational fluid dynamics, computational finance, seismic analysis, molecular modeling, and genomics.

Specification Name

vCPU

Memory    (GB)

GPU

Graphics  Memory (GB)

Virtualization Type

p2vs.4xlarge.8

16

128

1*V100s

1*32GB

KVM

p2vs.8xlarge.8

32

256

2*V100s

2*32GB

KVM

Common Software Compatibility List

The P2Vs model cloud server is primarily designed for compute-acceleration scenarios, such as deep learning training, inference, scientific computing, molecular modeling, seismic analysis, and other similar applications. If the application software utilizes the CUDA parallel computing capabilities of the GPU, the P2Vs model cloud server can be used. The common software support list is as follows:

  • Common deep learning frameworks:      TensorFlow, Caffe, PyTorch, MXNet

  • RedShift for Autodesk 3dsMax, V-Ray for      3ds Max, and other CUDA-enabled GPU rendering software

Tips for Use

The P2Vs model cloud server currently supports the following versions of operating systems:

  • Windows Server 2019 Datacenter 64bit

  • Windows Server 2016 Datacenter 64bit

  • CentOS 8.1 64bit

  • CentOS 7.8 64bit

  • Ubuntu Server 20.04 64bit

  • Ubuntu Server 18.04 64bit

PI2 model cloud server

The PI2 model cloud server is equipped with the NVIDIA Tesla T4 GPU, which is specifically designed for AI inference. With GPU pass-through technology, it can provide extremely powerful real-time inference capabilities. The PI2 model elastic cloud server, with the help of the T4's INT8 processor, can provide a maximum of 130

TOPS of INT8 computing power. The PI2 can also support lightweight training scenarios.

Specification    Name

vCPU

Memory    (GB)

GPU

Graphics    Memory (GB)

Virtualization Type

pi2.2xlarge.4

8

32

1×T4

1×16GB

KVM

pi2.4xlarge.4

16

64

2×T4

2×16GB

KVM

pi2.8xlarge.4

32

128

4×T4

4×16GB

KVM

Commonly Supported Software List

The PI2 model cloud server is primarily used for GPU inference computing scenarios, such as image recognition, speech recognition, natural language processing, and other similar applications. It can also support lightweight training scenarios.

The common software support list is as follows:

  • Common deep learning frameworks:      TensorFlow, Caffe, PyTorch, MXNet

  • RedShift for Autodesk 3dsMax, V-Ray for      3ds Max, and other CUDA-enabled GPU rendering software

Tips for Use

The PI2 model cloud server currently supports the following versions of operating systems:

  • Windows Server 2019 Datacenter 64bit

  • Windows Server 2016 Datacenter 64bit

  • CentOS 8.1 64bit

  • CentOS 7.8 64bit

  • Ubuntu Server 20.04 64bit

  • Ubuntu Server 18.04 64bit

PAK1 model cloud server

The PAK1 model Ascend GPU cloud server, equipped with the Atlas 300I Pro accelerator card specifically designed for AI inference, and powered by domestically produced ARM architecture Kunpeng CPU, belongs to the compute acceleration (pass-through) specification. The cloud server has a CPU/memory ratio of 1:4, primarily suitable for inference scenarios such as search recommendation, content audit, and OCR systems.

Series

CPU

Memory

GPU Graphics Card Type

Graphics Memory

Maximum Bandwidth/Base    Bandwidth (Gbit/s)

Number of Network Packages Sent    and Received (10,000 PPS)

Multi-queue

pak1.4xlarge.4

18

72

Huawei   Atlas 300I pro

1*24G

18/11.5

240

8

pak1.9xlarge.4

36

144

Huawei   Atlas 300I pro

2*24G

30/22.5

480

16

pak1.18xlarge.4

72

288

Huawei   Atlas 300I pro

4*24G

47/45

960

32

Supported Images

  • CTyunOS 2.0.1 64-bit ARM version

Note:   The PAK1 model GPU cloud server's public images all come with the CANN 3.1.0 development kit package pre-installed by default, and the environment variables have been configured.


GPU image acceleration basic type

The Image Acceleration Basic Type GPU cloud server is based on NVIDIA GRID virtualization GPU technology. The public image integrates the GRID driver and includes the software License for NVIDIA GRID vWS. It can effectively reduce the usage cost for small-scale demands and is suitable for scenarios such as image rendering and small-scale AI inference.

On Sale: G7

Image Acceleration Basic Type GPU Cloud Server features:

Specification Name

GPU Model

Quantity of GPUs

Single GPU Performance

Disk Type

G7

Nvidia   Tesla A10

1/41/21

31.2   TFLOPS Single Precision Floating Point Calculation
  62.5 TFLOPS TF32AI Acceleration

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

G6

Nvidia   Tesla T4

1/41/2

31.2   TFLOPS Single Precision Floating Point Calculation
  62.5 TFLOPS TF32AI Acceleration

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

G5

Nvidia   Tesla V100

1/161/81/41/2

14   TFLOPS Single Precision Floating Point Calculation
  7 TFLOPS Double Precision Floating Point Calculation
  112 TFLOPS TF32 AI Acceleration

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

G5s

Nvidia   Tesla V100s

1/161/81/41/2

16.4   TFLOPS Single Precision Floating Point Calculation
  8.2 TFLOPS Double Precision Floating Point Calculation
  130 TFLOPS Double Precision Floating Point Calculation

Standard   IO
  High IO
  General Purpose SSD
  Ultra High IO

G7 model cloud server

The G7 model cloud server, based on NVIDIA GRID virtualization GPU technology, is capable of providing comprehensive professional-level graphics acceleration capabilities. The G7 model cloud server uses the NVIDIA A10 Tensor Core GPU graphics card, capable of supporting DirectX, OpenGL, and Vulkan interfaces. It offers three video memory specifications of 6/12/24 GB and supports a resolution of 7680*4320, meeting the graphic processing needs from entry-level to professional-level.

Specification    Name

vCPU

Memory    (GB)

GPU

Graphics    Memory (GB)

Virtualization    Type

Maximum Bandwidth (Gbps) / Baseline Bandwidth (Gbps)

Number of NIC Queues

Maximum Packet Handling Capacity (in 10,000 PPS)

g7.2xlarge.4

8

32

A10-6Q

6

KVM

8/2.5

4

110

g7.4xlarge.4

16

64

A10-12Q

12

KVM

15/4.5

8

220

g7.8xlarge.4

32

128

A10-24Q

24

KVM

20/9

16

440

The G7 model cloud server features are as follows:

  • The processor-to-memory ratio is 1:4

  • Supports graphics acceleration      interfaces:

    • DirectX 12, Direct2D, DirectX Video       Acceleration(DXVA)

    • OpenGL 4.5

    • Vulkan 1.0

  • Supports CUDA and OpenCL

  • Supports Quadro vDWS features, providing acceleration for professional-grade graphics applications

  • Supports NVIDIA A10 GPU cards

  • Supports graphics acceleration applications

  • Provides GPU hardware virtualization  (vGPU)

  • Provides the same application process as elastic cloud servers.

  • Automates the scheduling of G7 model elastic cloud servers to available zones equipped with NVIDIA A10 GPU cards

  • It can provide graphic and image processing capabilities with a maximum video memory of 24GB and a resolution of 7680*4320

Commonly Supported Software List

The G7 model cloud server is primarily used for graphics acceleration scenarios, such as image rendering, cloud laptop, and 3D visualization. Application software that relies on the DirectX and OpenGL hardware acceleration capabilities of the GPU can utilize the G7 model cloud server. The common graphics processing software support list is as follows:

  • AutoCAD

  • 3DS MAX

  • MAYA

  • Agisoft PhotoScan

  • ContextCapture

Tips for Use

The G7 model cloud server currently supports the following versions of operating systems:

  • WindowsServer 2019 DataCenter 64bit

  • Windows Server 2016 DataCenter 64bit

  • Windows Server 2012 DataCenter 64bit

  • CentOS 8.1 64bit

  • CentOS 8.2 64bit

  • Ubuntu Server 20.04 64bit

G6 model cloud server

The G6 model cloud server, based on NVIDIA GRID virtualization GPU technology and equipped with the NVIDIA Tesla T4 GPU graphics card, supports DirectX, OpenGL, and Vulkan interfaces. It offers 8GiB of video memory and has a theoretical performance of a Pixel Rate of 101.8 GPixel/s and a Texture Rate of 254.4 GTexel/s, meeting the needs of professional-level graphics processing.

Specification Name

vCPU

Memory  (GB)

GPU

Graphics Memory (GB)

Virtualization Type

g6.xlarge.4

4

16

T4-4Q

4

KVM

g6.2xlarge.4

8

32

T4-8Q

8

KVM

Commonly Supported Software List

The G6 model cloud server is primarily used for graphics acceleration scenarios, such as image rendering, cloud laptop, and 3D visualization. Application software that relies on the DirectX and OpenGL hardware acceleration capabilities of the GPU can utilize the G6 model cloud server. The common graphics processing software support list is as follows:

  • AutoCAD

  • 3DS MAX

  • MAYA

  • Agisoft PhotoScan

  • ContextCapture

Tips for Use

The G6 model cloud server currently supports the following versions of operating systems:

  • Windows Server 2016 Standard 64bit

  • Windows Server 2012 Standard 64bit

  • CentOS 7.5 64bit

  • CentOS 7.6 64bit

  • Ubuntu Server 16.04 64bit

G5 model cloud server

The G5 model cloud server, based on NVIDIA GRID virtualization GPU technology, is capable of providing comprehensive professional-level graphics acceleration capabilities. The G5 model cloud server uses the NVIDIA Tesla V100 PCIe GPU graphics card, which supports DirectX, OpenGL, and Vulkan interfaces. It offers four video memory specifications: 2 GB, 4 GB, 8 GB, and 16 GB, and supports a maximum resolution of 4096*2160. This meets the graphic processing needs ranging from entry-level to professional-level.

Specification Name

vCPU

Memory (GB)

GPU

Graphics Memory (GB)

Virtualization Type

g5.2xlarge.2.1

8

16

V100-2Q

2

KVM

g5.2xlarge.2

8

32

V100-4Q

4

KVM

g5.2xlarge.8

8

64

V100-16Q

16

KVM

g5.4xlarge.4

16

64

V100-8Q

8

KVM

g5.8xlarge.4

32

128

V100-16Q

16

KVM

The G5 model cloud server features are as follows:

  • The processor-to-memory ratio is 1:4/1:2/1:8

  • Supports graphics acceleration  interfaces:

    • DirectX 12, Direct2D, DirectX Video  Acceleration(DXVA)

    • OpenGL 4.5

    • Vulkan 1.0

  • Supports CUDA and OpenCL

  • Supports Quadro vDWS features, providing acceleration for professional-grade graphics applications

  • Supports NVIDIA V100 GPU cards

  • Supports graphics acceleration applications

  • Provides GPU hardware virtualization  (vGPU)

  • Provides the same application process as elastic cloud servers.

  • Automates the scheduling of G5 model elastic cloud servers to available zones equipped with NVIDIA V100 GPU cards

  • It can provide graphic processing  capabilities with a maximum video memory of 16GB and a resolution of 4096×2160

Commonly Supported Software List

The G5 model cloud server is primarily used for graphics acceleration scenarios, such as image rendering, cloud laptop, and 3D visualization. Application software that relies on the DirectX and OpenGL hardware acceleration capabilities of the GPU can utilize the G5 model cloud server. The common graphics processing software support list is as follows:

  • AutoCAD

  • 3DS MAX

  • MAYA

  • Agisoft PhotoScan

  • ContextCapture

Tips for Use

The G5 model cloud server currently supports the following versions of operating systems:

  • Windows Server 2016 Standard 64bit

  • Windows Server 2012 Standard 64bit

  • CentOS 7.5 64bit

  • CentOS 7.6 64bit

  • Ubuntu Server 16.04 64bit

G5s model cloud server

The G5s model cloud server, based on NVIDIA GRID virtualization GPU technology, is capable of providing comprehensive professional-level graphics acceleration capabilities. The G5s model cloud server uses the NVIDIA Tesla V100s PCIe GPU graphics card, which supports DirectX, OpenGL, and Vulkan interfaces. It offers four video memory specifications: 2 GB, 4 GB, 8 GB, and 16 GB, and supports a maximum resolution of 4096*2160. This meets the graphic processing needs ranging from entry-level to professional-level.

Specification Name

vCPU

Memory (GB)

GPU

Graphics Memory (GB)

Virtualization Type

g5s.2xlarge.2.1

8

16

V100s-2Q

2

KVM

g5s.2xlarge.2

8

32

V100s-4Q

4

KVM

g5s.2xlarge.8

8

64

V100s-16Q

16

KVM

g5s.4xlarge.4

16

64

V100s-8Q

8

KVM

g5s.8xlarge.4

32

128

V100s-16Q

16

KVM

The G5s model cloud server features are as follows:

  • The processor-to-memory ratio is 1:4/1:2/1:8

  • Supports graphics acceleration interfaces:

    • DirectX 12, Direct2D, DirectX Video Acceleration(DXVA)

    • OpenGL 4.5

    • Vulkan 1.0

  • Supports CUDA and OpenCL

  • Supports Quadro vDWS features, providing acceleration for professional-grade graphics applications

  • Supports NVIDIA V100s GPU cards

  • Supports graphics acceleration applications

  • Provides GPU hardware virtualization (vGPU)

  • It offers the same application process as the elastic cloud server.

  • Automates the scheduling of G5s model elastic cloud servers to available zones equipped with NVIDIA V100s GPU cards

  • It can provide graphic processing capabilities with a maximum video memory of 16GB and a resolution of 4096×2160

Commonly Supported Software List

The G5s model cloud server is primarily used for graphics acceleration scenarios, such as image rendering, cloud laptop, and 3D visualization. Application software that relies on the DirectX and OpenGL hardware acceleration capabilities of the GPU can utilize the G5s model cloud server. The common graphics processing software support list is as follows:

  • AutoCAD

  • 3DS MAX

  • MAYA

  • Agisoft PhotoScan

  • ContextCapture

Tips for Use

The G5s model cloud server currently supports the following versions of operating systems:

  • Windows Server 2016 Standard 64bit

  • Windows Server 2012 Standard 64bit

  • CentOS 7.5 64bit

  • CentOS 7.6 64bit

  • Ubuntu Server 16.04 64bit


uOOSi3kluazi