Docs

GPU Compute Service

Instance types

GPU instance naming and type recommendations

The platform offers a variety of instance types to meet your requirements in different scenarios. This topic describes how to select and purchase a GPU instance based on business requirements to empower your businesses with the elastic ECS service.

Instance name

The GPU Compute Service (GPU) servers offered by the platform can be categorized into Compute-optimized and HPC-optimized based on performance.

Instance Families

Format: <Instance family series><GPU graphics card type><Processor type><Instance family generation><Instance family suffix>

Instance type family series	GPU graphics card type	Processor Type	Instance type family generation	Suffix
Lowercase letters, representing an instance type family type, are usually abbreviations of certain words that represent the performance domain of the instance type family. p: Performance, general performance type t: Training, applicable for training scenarios i: Inference, applicable for inference scenarios g: Graphic, graphics-intensive ebm prefix: the EBM type hpc prefix: HPC-optimized instances with access to RDMA network	A lowercase English letter representing the GPU graphics card vendor. n：NVIDIA GPU	A lowercase English letter representing the CPU vendor. i: Intel a: AMD r: ARM	Arabic numerals that indicate the generation of the instance type family. A larger number indicates a later generation.	Lowercase English letters, indicating other attributes of the instance family c: customized, indicating customized edition l: light, indicating the lightweight edition e: balanced enhanced attribute. For example, balanced vCPU, memory, and network capability. h: high-performance, indicating high-performance edition d: equipped with local volume v/t: GPU card of V/T series vws: Rendering instance configured with vDWS License server and GRID driver For more instance attributes, please refer to Instance types.

Instance type

Format: ecs.<Instance family name><Processor tag>.<Size>

ecs	Instance type family name	Size
Represents the product name Elastic Compute Service (ECS) and cannot be changed.	For details, refer to Instance type family name.	The number of vCPUs for an instance type, typically represented as large or <nx>large. large represents 2 vCPUs, xlarge represents 4 vCPUs, 2xlarge represents 2 * 4 = 8 vCPUs, and 3xlarge represents 3 * 4 = 12 vCPUs. So, a larger n value represents more vCPUs and a larger specification. For example, ecs.hpcpni2.28xlarge represents an instance in the HPC-optimized pni2 instance family with 112 vCPUs.

Instance type

ECS offers multiple instance types with different computing and storage capabilities to suit different scenarios. For more information on instance types, refer to Instance types.

Due to the multiple types of instance families, some instance families do not strictly follow the above naming rules. Details are as follows:

Instance family type	Identifier for instance family	Description
GPU Accelerated	ini2	i: equipped with inference performance GPU card. n: the NVIDIA GPU graphics card. i: the Intel processor. 2: the 2nd-generation series.
		pni2 pni3l/pni3ld	p: general performance GPU card. n: the NVIDIA GPU graphics card. i: the Intel processor. 2/3: instance generation.
		gni2	g: Graphic, the GPU card applicable for graphics-intensive scenarios. n: The NVIDIA GPU graphics card. i: the Intel processor. 2/3: instance generation.
		g1ve g1vc g1te g1tl	g: the GPU card. t: The GPU card type is T4. You can mount a maximum of 8 T4 cards for one instance. v: GPU card type is V100. You can mount a maximum of 8 V100 cards for one instance. l: the streamlined attribute. You can mount a maximum of 4 T4 cards for one instance. e: the enhanced balance attribute. c: customized, indicating customized edition. 1: the 1st-generation series.
		gni3 ebmgni3 ebmgni3c gni3c gni3cg/gni3cgd gni3cl gna3c/gna3cd	ebm: The EBM server. g: Graphic, the GPU card applicable for graphics-intensive scenarios. n: The NVIDIA GPU graphics card. i: the Intel processor. a: the AMD processor. 3: the 3rd-generation series. d: Local volumes of SSD type are offered.
GPU Rendering	g1te-vws	g: Graphic, the GPU card applicable for graphics-intensive scenarios. t: The GPU card type is T4. You can mount a maximum of one T4 cards. e: the enhanced balance attribute. 1: the 1st-generation series. vws: configured with vDWS License server and GRID driver installed.
		gni2-vws	g: Graphic, the GPU card applicable for graphics-intensive scenarios. n: The NVIDIA GPU graphics card. i: the Intel processor. 2: the 2nd-generation series. vws: configured with vDWS License server and GRID driver installed
		gni3l-vws gni3cg-vws gni3cl-vws	g: Graphic, the GPU card applicable for graphics-intensive scenarios. n: The NVIDIA GPU graphics card. i: Intel processors are associated. 3: the 3rd-generation series. vws: configured with vDWS License server and GRID driver installed
HPC-optimized	hpcpni2	hpc: HPC-optimized instances that support RDMA network. p: general performance GPU cards are associated. n: the NVIDIA GPU graphics card. i: the Intel processor. 2: the 2nd-generation series. It includes a local volume of SSD type.
		hpcg1ve	hpc: HPC-optimized instances that support RDMA network. g: GPU card associated. v: GPU card type is V100. e: the enhanced balance attribute. 1: the 1st-generation series.
		ebmhpcpni2l ebmhpcpni2 ebmhpchfpni2	hpc: HPC-optimized instances that support RDMA network. ebm: The EBM server. hf: High Frequency p: general performance GPU card. n: the NVIDIA GPU graphics card. i: the Intel processor. 1: the 2nd-generation series. l: light, indicating the lightweight edition. It includes a local volume of SSD type.
		ebmhpcpni3l hpcpni3ln hpcpni3le	ebm: Elastic Bare Metal. hpc: HPC-optimized instances that support RDMA network. p: general performance GPU cards are associated. n: the NVIDIA GPU graphics card. i: Intel processors are associated. 3: the 3rd-generation series. n: Network. l: light, indicating the lightweight edition. e: Enhance, indicating enhanced attribute.

Applicable scenarios

Instance type\Applicable scenario	Scientific computing	Deep learning inference	Deep learning training	Graphic rendering	Video encoding
GPU Accelerated g1ve	Recommended	Recommended	Recommended	Supported	Supported
GPU Accelerated g1vc	Recommended	Recommended	Recommended	Supported	Supported
GPU Accelerated g1te	Not supported	Recommended	Supported	Recommended	Supported
GPU Accelerated g1tl	Not supported	Recommended	Supported	Recommended	Supported
GPU Accelerated gni2	Supported	Recommended	Recommended	Supported	Supported
GPU Accelerated ini2	Recommended	Supported	Recommended	Not supported	Not supported
GPU Accelerated pni2	Recommended	Supported	Recommended	Not supported	Not supported
GPU Accelerated pni3l	Supported	Recommended	Recommended	Not supported	Not supported
GPU Accelerated gni3/ebmgni3	Supported	Recommended	Not supported	Recommended	Recommended
GPU Accelerated gni3cgd/gni3c/ebmgni3c/gni3cg/gni3cl/gna3c	Recommended	Recommended	Supported	Not supported	Not supported
GPU Rendering gni3cl-vws/gni3cg-vws/gni3l-vws	Not supported	Supported	Not supported	Recommended	Recommended
GPU Rendering g1te-vws	Not supported	Supported	Supported	Recommended	Recommended
GPU Rendering gni2-vws	Not supported	Supported	Supported	Recommended	Recommended
HPC-optimized hpcpni2	Recommended	Supported	Recommended	Not supported	Not supported
HPC-optimized hpcg1ve	Recommended	Recommended	Recommended	Supported	Supported
HPC-optimized ebmhpcpni2	Recommended	Supported	Recommended	Not supported	Not supported
HPC-optimized ebmhpchfpni2	Recommended	Supported	Recommended	Not supported	Not supported
HPC-optimized ebmhpcpni2l	Recommended	Supported	Recommended	Not supported	Not supported
HPC-optimized ebmhpcpni3l	Supported	Recommended	Recommended	Not supported	Not supported
HPC-optimized hpcpni3ln	Supported	Recommended	Recommended	Not supported	Not supported
HPC-optimized hpcpni3le	Supported	Recommended	Recommended	Not supported	Not supported
HPC-optimized hpcpni3h	Recommended	Recommended	Supported	Not supported	Not supported

Recommended instance type families for common business scenarios

Business scenario	Commonly used software	Recommended instance type
Deep learning	MXNet, TensorFlow, PyTorch	GPU Accelerated pni3l/pni2/g1ve/g1vc, HPC-optimized hpcg1ve
AI training	NXNET	GPU Accelerated gni2/ini2/pni2/pni3l/gni3cl, HPC-optimized
AI Inference	OpenVINO, TensorRT	GPU Accelerated gni2/ini2/g1ve/g1vc, HPC-optimized ebmhpcpni3l/hpcpni3ln/hpcpni3h
Graphic/video rendering	Unity, UE4	GPU Rendering
HPC cluster computing training	StarCCM, WRF-Chem	HPC-optimized

Last updated: 2026.06.24 18:25:48