vGPU-accelerated instance families (vgn and sgn series)

Updated at:

This page describes the features and instance types of the following vGPU-accelerated instance families in Elastic GPU Service (EGS):

  • sgn8ia: Latest-generation vGPU instances powered by NVIDIA Lovelace GPUs and AMD Genoa processors

  • sgn7i-vws: Cost-efficient vGPU instances with shared CPUs and NVIDIA A10 GPUs

  • vgn7i-vws: Dedicated-CPU vGPU instances with NVIDIA A10 GPUs

  • vgn6i-vws: Previous-generation vGPU instances with NVIDIA T4 GPUs (upgraded from vgn6i)

All families run on the third-generation SHENLONG architecture with fast path acceleration, delivering storage, network, and compute stability improvements of an order of magnitude over traditional virtualization. Each family includes an NVIDIA GRID virtual workstation (vWS) license, providing certified graphics acceleration for Computer Aided Design (CAD) software and professional graphics applications.

Family comparison

Family

GPU

CPU

CPU allocation

vGPU range

Storage

sgn8ia

NVIDIA Lovelace

AMD Genoa, 3.4–3.75 GHz

Shared (~1:1.5 overcommit)

2 GB to 48 GB GPU memory

ESSDs, ESSD AutoPL disks

sgn7i-vws

NVIDIA A10 (Ampere)

Intel Xeon Ice Lake, 2.9/3.5 GHz

Shared

1/12 to 1/3 of A10 (2–8 GB GPU memory)

ESSDs, ESSD AutoPL disks

vgn7i-vws

NVIDIA A10 (Ampere)

Intel Xeon Ice Lake, 2.9/3.5 GHz

Dedicated

1/6 to full A10 (4–24 GB GPU memory)

ESSDs, ESSD AutoPL disks

vgn6i-vws

NVIDIA T4

Intel Xeon Platinum 8163 (Skylake), 2.5 GHz

Dedicated

1/4 to full T4 (4–16 GB GPU memory)

Standard SSDs, ultra disks

How vGPU slicing works

Each physical GPU is sliced into multiple GPU partitions. Each partition is allocated as a vGPU to a single instance. The GPUs column in the instance type tables uses the format <GPU model> * <fraction> to show both the GPU model and the partition size allocated to each instance. For example, NVIDIA A10 * 1/6 means the instance receives one-sixth of an NVIDIA A10 GPU.

Use cases

Use case

Description

Recommended families

Remote graphics and virtual workstations

Graphic design, CAD, animation, film production, mechanical design — accessed remotely with near-native GPU performance

sgn8ia, sgn7i-vws, vgn7i-vws

AI inference at scale

Concurrent inference for image recognition, speech recognition, and behavior identification

sgn8ia, sgn7i-vws, vgn7i-vws

Cloud gaming

Real-time GPU rendering for interactive cloud gaming and AR/VR applications

All families

3D visualization

Professional-grade GPU rendering for graphics-intensive workloads

sgn8ia, sgn7i-vws, vgn7i-vws

Deep learning environments

Educational and experimental deep learning environments requiring GPU acceleration

vgn6i-vws

sgn8ia

sgn8ia instances use NVIDIA Lovelace GPUs with large GPU memory and multiple GPU slicing options, paired with AMD Genoa processors running at 3.4 GHz to 3.75 GHz. CPUs are shared resources with an overcommit ratio of approximately 1:1.5. Memory and GPU memory are exclusive to each instance. Available GPU memory ranges from 2 GB to 48 GB (full GPU).

For workloads that require dedicated CPUs, use gn7i GPU-accelerated compute-optimized instances instead.

GPU: NVIDIA Lovelace — supports vGPU, RTX, and TensorRT

CPU: AMD Genoa, 3.4 GHz to 3.75 GHz (shared, ~1:1.5 overcommit)

Storage: I/O optimized; supports the NVMe protocol. For more information, see NVMe protocol. Supports Enterprise SSDs (ESSDs), ESSD AutoPL disks, and regional Enterprise SSDs (ESSDs).

Network: Supports IPv4 and IPv6

Use cases:

  • Concurrent AI inference — image recognition, speech recognition, behavior identification

  • Compute-intensive graphics processing — remote graphic design, cloud gaming

  • 3D modeling — animation, film production, cloud gaming, mechanical design

Instance types

Instance type

vCPUs

Memory (GiB)

GPU memory

Network baseline bandwidth (Gbit/s)

Packet forwarding rate (pps)

NIC queues

ENIs

Private IPv4/IPv6 addresses per ENI

Maximum disks

Disk baseline IOPS

Disk baseline BPS (MB/s)

ecs.sgn8ia-m2.xlarge

4

16

2 GB

2.5

1,000,000

4

4

15/15

9

30,000

244

ecs.sgn8ia-m4.2xlarge

8

32

4 GB

4

1,600,000

8

4

15/15

9

45,000

305

ecs.sgn8ia-m8.4xlarge

16

64

8 GB

7

2,000,000

16

8

30/30

17

60,000

427

ecs.sgn8ia-m16.8xlarge

32

128

16 GB

10

3,000,000

32

8

30/30

33

80,000

610

ecs.sgn8ia-m24.12xlarge

48

192

24 GB

16

4,500,000

48

8

30/30

33

120,000

1,000

ecs.sgn8ia-m48.24xlarge

96

384

48 GB

32

9,000,000

64

15

30/30

33

24,000

2,000

GPU memory values represent vGPU memory allocated using vGPU slicing technology. CPUs are shared with an overcommit ratio of approximately 1:1.5. Memory and GPU memory are exclusive to each instance.

sgn7i-vws

sgn7i-vws instances use NVIDIA A10 GPUs (NVIDIA Ampere architecture) paired with Intel Xeon Scalable processors (Ice Lake, 2.9 GHz base, 3.5 GHz all-core turbo). CPU and network resources are shared to maximize utilization. Memory and GPU memory are exclusive to each instance. Available GPU memory ranges from 1/12 of an A10 GPU (2 GB) to 1/3 of an A10 GPU (8 GB).

For workloads that require dedicated CPUs, use vgn7i-vws instead.

GPU: NVIDIA A10 (Ampere architecture) — supports vGPU, RTX, and TensorRT

CPU: Intel Xeon Scalable (Ice Lake), 2.9 GHz base / 3.5 GHz all-core turbo (shared)

Storage: I/O optimized; supports ESSDs and ESSD AutoPL disks

Network: Supports IPv4 and IPv6 (shared)

Use cases:

  • Concurrent AI inference — image recognition, speech recognition, behavior identification

  • Compute-intensive graphics processing — remote graphic design, cloud gaming

  • 3D modeling — animation, film production, cloud gaming, mechanical design

Instance types

Instance type

vCPUs

Memory (GiB)

GPUs

GPU memory

Network baseline/burst bandwidth (Gbit/s)

Packet forwarding rate (pps)

NIC queues

ENIs

Private IPv4 addresses per ENI

IPv6 addresses per ENI

ecs.sgn7i-vws-m2.xlarge

4

15.5

NVIDIA A10 * 1/12

24GB * 1/12

1.5/5

500,000

4

2

2

1

ecs.sgn7i-vws-m4.2xlarge

8

31

NVIDIA A10 * 1/6

24GB * 1/6

2.6/10

1,000,000

4

4

6

1

ecs.sgn7i-vws-m8.4xlarge

16

62

NVIDIA A10 * 1/3

24GB * 1/3

5/20

2,000,000

8

4

10

1

ecs.sgn7i-vws-m2s.xlarge

4

8

NVIDIA A10 * 1/12

24GB * 1/12

1.5/5

500,000

4

2

2

1

ecs.sgn7i-vws-m4s.2xlarge

8

16

NVIDIA A10 * 1/6

24GB * 1/6

2.5/10

1,000,000

4

4

6

1

ecs.sgn7i-vws-m8s.4xlarge

16

32

NVIDIA A10 * 1/3

24GB * 1/3

5/20

2,000,000

8

4

10

1

The GPUs column shows the GPU model and partition allocated per instance. NVIDIA A10 * 1/12 means each instance receives one-twelfth of an NVIDIA A10 GPU as a vGPU. CPU and network resources are shared; memory and GPU memory are exclusive to each instance.

vgn7i-vws

vgn7i-vws instances use NVIDIA A10 GPUs (NVIDIA Ampere architecture) paired with Intel Xeon Scalable processors (Ice Lake, 2.9 GHz base, 3.5 GHz all-core turbo). Unlike sgn7i-vws, CPU resources are dedicated to each instance. Available GPU memory ranges from 1/6 of an A10 GPU (4 GB) to a full A10 GPU (24 GB).

GPU: NVIDIA A10 (Ampere architecture) — supports vGPU, RTX, and TensorRT

CPU: Intel Xeon Scalable (Ice Lake), 2.9 GHz base / 3.5 GHz all-core turbo (dedicated)

Storage: I/O optimized; supports ESSDs and ESSD AutoPL disks

Network: Supports IPv4 and IPv6

Use cases:

  • Concurrent AI inference — image recognition, speech recognition, behavior identification

  • Compute-intensive graphics processing — remote graphic design, cloud gaming

  • 3D modeling — animation, film production, cloud gaming, mechanical design

Instance types

Instance type

vCPUs

Memory (GiB)

GPUs

GPU memory

Network baseline bandwidth (Gbit/s)

Packet forwarding rate (pps)

NIC queues

ENIs

Private IPv4 addresses per ENI

IPv6 addresses per ENI

ecs.vgn7i-vws-m4.xlarge

4

30

NVIDIA A10 * 1/6

24GB * 1/6

3

1,000,000

4

4

10

1

ecs.vgn7i-vws-m8.2xlarge

10

62

NVIDIA A10 * 1/3

24GB * 1/3

5

2,000,000

8

6

10

1

ecs.vgn7i-vws-m12.3xlarge

14

93

NVIDIA A10 * 1/2

24GB * 1/2

8

3,000,000

8

6

15

1

ecs.vgn7i-vws-m24.7xlarge

30

186

NVIDIA A10 * 1

24GB * 1

16

6,000,000

12

8

30

1

The GPUs column shows the GPU model and partition allocated per instance. NVIDIA A10 * 1/6 means each instance receives one-sixth of an NVIDIA A10 GPU as a vGPU. CPU resources are dedicated; memory and GPU memory are exclusive to each instance.

vgn6i-vws

Important

vgn6i-vws is the upgraded version of vgn6i, updated to use the latest NVIDIA GRID driver with an NVIDIA GRID vWS license.

  • Free images with the driver pre-installed: Submit a ticket to request a pre-installed image.

  • Custom images without the driver: Submit a ticket to apply for the driver file. Alibaba Cloud does not charge additional license fees.

vgn6i-vws instances use NVIDIA T4 GPUs paired with Intel Xeon Platinum 8163 processors (Skylake, 2.5 GHz). The CPU-to-memory ratio is 1:5. Supports 1/4 and 1/2 compute capacity of NVIDIA Tesla T4 GPUs, with 4 GB and 8 GB of GPU memory per vGPU instance.

GPU: NVIDIA T4 — supports 1/4 and 1/2 compute capacity, 4 GB and 8 GB GPU memory per vGPU instance

CPU: Intel Xeon Platinum 8163 (Skylake), 2.5 GHz

Storage: I/O optimized; supports standard SSDs and ultra disks

Network: Supports IPv4 and IPv6

Use cases:

  • Real-time rendering for cloud gaming

  • Real-time rendering for Augmented Reality (AR) and Virtual Reality (VR) applications

  • AI inference — deep learning and machine learning for elastic internet service deployment

  • Deep learning educational and experimental environments

Instance types

Instance type

vCPUs

Memory (GiB)

GPUs

GPU memory

Network baseline bandwidth (Gbit/s)

Packet forwarding rate (pps)

NIC queues

ENIs

Private IPv4 addresses per ENI

IPv6 addresses per ENI

ecs.vgn6i-m4-vws.xlarge

4

23

NVIDIA T4 * 1/4

16GB * 1/4

2

500,000

4/2

3

10

1

ecs.vgn6i-m8-vws.2xlarge

10

46

NVIDIA T4 * 1/2

16GB * 1/2

4

800,000

8/2

4

10

1

ecs.vgn6i-m16-vws.5xlarge

20

92

NVIDIA T4 * 1

16GB * 1

7.5

1,200,000

6

4

10

1

The GPUs column shows the GPU model and partition allocated per instance. NVIDIA T4 * 1/4 means each instance receives one-quarter of an NVIDIA T4 GPU as a vGPU.