Docs

Volcengine Kubernetes Engine

Instance families

GPU-accelerated specifications

GPU-accelerated VCIs use various NVIDIA Tesla graphics cards and deliver cost-effective performance across diverse inference scenarios and molecular computing use cases.

Refer to the introduction below to select a GPU-accelerated VCI specification that matches your actual business needs.

Instance family	Description
GPU-accelerated gni3 (vci.gni3)	(Beta) This instance family uses NVIDIA GPUs and is ideal for generative AI inference scenarios with deployment at scale. It supports NVIDIA RTX™ technology and provides high-performance 3D graphics virtualization capabilities.
GPU-accelerated gni2 (vci.gni2)	(Available in limited quantities) This instance family uses A10 graphics cards and is suitable for deep learning inference scenarios and small-scale training scenarios. It supports NVIDIA RTX™ technology and delivers high-performance 3D graphics virtualization capabilities.
GPU-accelerated ini2 (vci.ini2)	(Available in limited quantities) This instance family uses A30 graphics cards and supports powerful double-precision floating-point computing and high-throughput deep learning inference. It is applicable to large-scale AI inference and small-scale deep learning training. It does not support image or video rendering.
GPU-accelerated g1v (vci.g1v)	(Available in limited quantities) This instance family uses V100 graphics cards. It is suitable for:- Deep learning-oriented AI algorithm training applications (e.g., image classification, autonomous driving, speech recognition) - Scientific computing scenarios (e.g., computational fluid dynamics, computational finance, molecular dynamics, environmental analysis)

Tip

Supported specifications vary by regions and availability zones. For details, see Regions and availability zones.

GPU-accelerated gni3

Tip

[Beta - apply for trial]: This instance family is in the beta phase. To use it, submit an application.

Features

Type	Performance
Computing	Uses 4th generation Intel® Xeon® scalable processors (Sapphire Rapids). NVIDIA GPUs with 24 GB of video memory per GPU are used. Each instance supports up to 4 GPUs.
Storage	Supports ESSD cloud disks, with a maximum of 16 cloud disks mountable (including one system disk). Storage I/O performance is correlated with instance specifications: a higher specification delivers stronger performance.
Network	Network performance is correlated with instance specifications: a higher specification delivers stronger performance.

Application scenarios

This instance family delivers excellent AI video performance that is 120 times higher than that of CPU-based video performance solutions, making it suitable for the following scenarios:

Large-scale generative AI inference.

Accelerated video, AI, and graphics workloads.

Supports NVIDIA RTX™ features, providing high-performance 3D graphics virtualization capabilities. It is applicable to the following scenarios:

Graphics and image processing.

Video encoding and decoding.

Graph databases.

Usage method

For instructions on using GPU-accelerated gni3 instances, see Creating GPU-accelerated VCIs with a specified instance specification and Creating GPU-accelerated instances with specified GPU resources for containers.

Specifications list

Instance specification	vCPU	Memory (GiB)	GPU quantity	GPU memory (GB)
vci.gni3.12c-64gi	12	62	GPU × 1	24 × 1
vci.gni3.24c-128gi	24	124	GPU × 2	24 × 2
vci.gni3.48c-256gi	48	248	GPU × 4	24 × 4

GPU-accelerated gni2

Tip

Instances with these specifications are available in limited quantities. To use these instances, submit a ticket.

Features

Type	Performance
Computing	Uses 3rd generation Intel® Xeon® scalable processors (Ice Lake) with a base frequency of 2.3 GHz and a maximum all-core turbo frequency of 3.0 GHz. The processor-to-memory ratio is 1:4. It offers up to 112 vCPUs and 448 GiB of memory. NVIDIA A10 GPUs with 24 GB of video memory per GPU are used. Each instance supports up to 4 GPUs.
Storage	ESSD cloud disks. Storage I/O performance is correlated with instance specifications: a higher specification delivers stronger performance.
Network	Maximum network bandwidth: 80 Gbit/s. Maximum packets per second: 10 million PPS. Network performance is correlated with instance specifications: a higher specification delivers stronger performance.

Application scenarios

It is applicable to inference and small-scale training scenarios in deep learning, for example:

AI inference with large-scale deployment.

Small-scale training.

Supports NVIDIA RTX™ features, providing high-performance 3D graphics virtualization capabilities. It is applicable to the following scenarios:

Graphics and image processing.

Video encoding and decoding.

Graph databases.

Usage method

For instructions on using GPU-accelerated gni2 instances, see Creating GPU-accelerated VCIs with a specified instance specification and Creating GPU-accelerated instances with specified GPU resources for containers.

Specifications list

Instance specification	vCPU	Memory (GiB)	GPU quantity	GPU Memory (GB)
vci.gni2.26c-112gi	26	112	A10 × 1	24 × 1
vci.gni2.54c-224gi	54	224	A10 × 2	24 × 2
vci.gni2.108c-448gi	108	448	A10 × 4	24 × 4

GPU-accelerated ini2

Tip

Instances with these specifications are available in limited quantities. If necessary, submit a ticket.

Features

Type	Performance
Computing	Uses 3rd generation Intel® Xeon® scalable processors (Ice Lake) with a base frequency of 2.3 GHz and a maximum all-core turbo frequency of 3.0 GHz. The processor-to-memory ratio is 1:9. It offers up to 108 vCPUs and 972 GiB of memory. NVIDIA A30 GPUs with 24 GB of video memory per GPU are used. Each instance supports up to 4 GPUs.
Storage	ESSD cloud disks. Storage I/O performance is correlated with instance specifications: a higher specification delivers stronger performance.
Network	Maximum network bandwidth: 80 Gbit/s. Maximum packets per second: 10 million PPS. Network performance is correlated with instance specifications: a higher specification delivers stronger performance.

Application scenarios

AI inference with large-scale deployment.

Small-scale training for deep learning, image recognition, speech recognition, and more.

Image and video rendering is not supported.

Usage method

For instructions on using GPU-accelerated ini2 instances, see Creating GPU-accelerated VCIs with a specified instance specification and Creating GPU-accelerated instances with specified GPU resources for containers.

Specifications list

Instance specification	vCPU	Memory (GiB)	GPU quantity	GPU memory (GB)
vci.ini2.26c-243gi	26	243	A30 × 1	24 × 1
vci.ini2.52c-486gi	52	486	A30 × 2	24 × 2
vci.ini2.108c-972gi	108	972	A30 × 4	24 × 4

GPU-Accelerated g1v

Tip

Instances with these specifications are available in limited quantities. If necessary, submit a ticket.

Features

Type	Performance
Computing	Uses 2nd Gen Intel® Xeon® Scalable Xeon Platinum 8260 processors (Cascade Lake) with a base frequency of 2.4 GHz and all-core turbo frequency of 3.1 GHz. The processor-to-memory ratio is 1:4. Each NVIDIA Tesla V100 GPU is equipped with 32 GB of GPU memory. Each instance supports up to 8 GPUs.
Storage	Supports ESSD cloud disks, with a maximum of 16 cloud disks mountable (including one system disk). Storage I/O performance is correlated with instance specifications: a higher specification delivers stronger performance.
Network	Network performance is correlated with instance specifications: a higher specification delivers stronger performance.

Application scenarios

Deep learning, for example, is used for training artificial intelligence algorithms such as image classification, autonomous driving, and speech recognition.

Scientific computing, for example, computational fluid dynamics, computational finance, molecular dynamics, and environmental analysis.

Usage method

For instructions on using GPU-accelerated g1v instances, see Creating GPU-accelerated VCIs with a specified instance specification and Creating GPU-accelerated instances with specified GPU resources for containers.

Specifications list

Instance specification	vCPU	Memory (GiB)	GPU quantity	GPU memory (GB)
vci.g1v.2xlarge	8	32	Tesla V100 × 1	32 × 1
vci.g1v.4xlarge	16	64	Tesla V100 × 2	32 × 2
vci.g1v.8xlarge	32	128	Tesla V100 × 4	32 × 4
vci.g1v.20xlarge	80	320	Tesla V100 × 8	32 × 8

Last updated: 2026.07.24 16:18:02