Alibaba Cloud GPU ECS instances (gn/vgn/sgn series) - Elastic Compute Service

Note

View instance availability by region: Instance types may vary by region. We recommend that you check the purchase availability in each region.
View instance type selection guide: First, determine which instance families are suitable for your business scenario. Then, use this topic to select a specific instance type.
View instance metric descriptions: Read this topic to understand the metrics for instance types.
Use the ECS Price Calculator: You can use the price calculator to estimate instance fees.

sgn8ia vGPU-accelerated instance family

Overview:
- Powered by the third-generation SHENLONG architecture, sgn8ia instances deliver stable and predictable ultra-high performance. Chip-level acceleration provides a significant boost in storage, network, and compute performance, helping you store data and load models faster.
- These instances include an NVIDIA GRID Virtual Workstation (vWS) software license that provides certified graphics acceleration for a variety of professional computer-aided design (CAD) applications. They can also serve as cost-effective, lightweight GPU-accelerated instances for small-scale AI reasoning workloads.
Use cases:
- Concurrent AI reasoning workloads that require high-frequency CPUs, ample memory, and powerful GPUs, such as image recognition, speech recognition, and behavior identification.
- Graphics-intensive applications that require high-performance 3D graphics virtualization, such as remote graphic design and cloud gaming. These instances support RTX features and are paired with high-frequency CPUs.
- 3D modeling for film and animation production, cloud gaming, and mechanical design. The high-frequency AMD Genoa processors, with clock speeds up to 3.75 GHz, provide superior performance for these tasks.
Compute:
- Equipped with NVIDIA Lovelace architecture GPUs that feature:
  - Large GPU memory and multiple GPU slicing options.
  - Support for common acceleration features such as vGPU, RTX, and TensorRT for a wide range of workloads.
- Processor: High-frequency AMD Genoa processors with clock speeds from 3.4 GHz to 3.75 GHz, delivering higher compute power for 3D modeling.
Storage:
- All instances in this family are I/O optimized.
- These instances support the NVMe protocol. For more information, see Overview of the NVMe protocol.
- Supported cloud disk types: ESSD cloud disks, ESSD AutoPL cloud disks, and regional ESSD cloud disks. For more information about cloud disks, see Block storage overview.
Network:
- These instances support IPv4 and IPv6. For more information, see IPv6 communication.
- Network performance scales with the instance type, as larger instances offer better performance.

The following table lists the instance types and specifications for the sgn8ia family.

Instance type	vCPUs	Memory (GiB)	GPU memory	Network bandwidth (Gbit/s)	Packet rate (pps)	NIC queues	ENIs	Private IPs per ENI	Maximum cloud disks	Baseline IOPS	Baseline throughput (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	1000
ecs.sgn8ia-m48.24xlarge	96	384	48 GB	32	9,000,000	64	15	30/30	33	240,000	2000

Note

The GPU specifications in the table refer to vGPU slices created using vGPU technology.
For sgn8ia instances, the memory and GPU memory are dedicated resources. The vCPUs are shared resources with an oversubscription ratio of approximately 1:1.5. If your workload requires dedicated CPU computing power, use a dedicated instance with passthrough GPUs, such as an instance from the gn7i GPU-accelerated compute-optimized family.

sgn7i-vws, vGPU-accelerated instance family with shared CPUs

Overview:
- Powered by the third-generation SHENLONG architecture, sgn7i-vws instances deliver stable, predictable, and high performance. They use chip-level fast path acceleration to improve storage, network performance, and compute stability, helping you store data and load models faster.
- Instances in this instance family share CPU and network resources to maximize underlying resource utilization. Memory and GPU memory are dedicated to each instance, ensuring data isolation and consistent performance.
  
  Note
  If you need dedicated CPU resources, select the vgn7i-vws instance family.
- These instances include an NVIDIA GRID Virtual Workstation (vWS) software license, which provides certified graphics acceleration drivers for various professional CAD applications. They can also be used as lightweight, cost-effective GPU compute instances for small-scale AI inference workloads.
Use cases:
- Equipped with high-performance CPUs, memory, and GPUs, these instances are suitable for concurrent AI inference tasks such as image recognition, speech recognition, and behavior recognition.
- Supporting RTX features and high-frequency CPUs, these instances provide high-performance 3D graphics virtualization, ideal for graphics-intensive workloads such as remote graphic design and cloud gaming.
- Powered by Intel Ice Lake processors, these instances deliver excellent performance for 3D modeling in fields such as film and animation production, cloud gaming, and mechanical design.
Compute:
- Equipped with NVIDIA A10 GPUs.
  - Innovative NVIDIA Ampere architecture.
  - Support for common acceleration features such as vGPU, RTX, and TensorRT for a variety of workloads.
- Processor: 2.9 GHz Intel^® Xeon^® Scalable (Ice Lake) processor with an all-core turbo frequency of 3.5 GHz.
Storage:
- I/O optimized instances.
- Supported cloud disk types: ESSD cloud disk, ESSD AutoPL cloud disk, and Zone-redundant ESSD cloud disk. For more information, see Block storage overview.
Network:
- Supports IPv4 and IPv6. For more information about IPv6 communication, see IPv6 communication.
- Network performance scales with the instance type; larger types offer higher performance.

The sgn7i-vws instance family includes the instance types and specifications listed in the following table:

Instance type	vCPU	Memory (GiB)	GPU	GPU memory	Network bandwidth (Gbit/s)	Packet rate (PPS)	NIC queues	ENIs	Private IPv4 addresses	IPv6 addresses
ecs.sgn7i-vws-m2.xlarge	4	15.5	NVIDIA A10 * 1/12	24 GB * 1/12	1.5/5	500,000	4	2	2	1
ecs.sgn7i-vws-m4.2xlarge	8	31	NVIDIA A10 * 1/6	24 GB * 1/6	2.6/10	1,000,000	4	4	6	1
ecs.sgn7i-vws-m8.4xlarge	16	62	NVIDIA A10 * 1/3	24 GB * 1/3	5/20	2,000,000	8	4	10	1
ecs.sgn7i-vws-m2s.xlarge	4	8	NVIDIA A10 * 1/12	24 GB * 1/12	1.5/5	500,000	4	2	2	1
ecs.sgn7i-vws-m4s.2xlarge	8	16	NVIDIA A10 * 1/6	24 GB * 1/6	2.6/10	1,000,000	4	4	6	1
ecs.sgn7i-vws-m8s.4xlarge	16	32	NVIDIA A10 * 1/3	24 GB * 1/3	5/20	2,000,000	8	4	10	1

Note

The GPU column shows the GPU model and GPU slicing information. GPU slicing divides a physical GPU into multiple slices and allocates one slice to each instance.

For example, in NVIDIA A10 * 1/12, NVIDIA A10 is the GPU model, and 1/12 means that a single physical GPU is divided into 12 slices, with one slice allocated to the instance.

vgn7i-vws, vGPU-accelerated instance family

Overview
- Built on the third-generation Shenlong architecture, these instances deliver stable and predictable high performance. They use chip-level fast path acceleration to significantly improve storage performance, network performance, and compute stability, allowing you to store data and load models faster.
- These instances include an NVIDIA GRID vWS software license, providing certified graphics acceleration drivers for professional CAD applications and graphic design workloads. They can also be used as cost-effective, lightweight compute instances for small-scale AI inference.
Use cases
- With high-performance CPUs, memory, and GPUs, these instances can handle a high volume of concurrent AI inference tasks, such as image recognition, speech recognition, and behavior recognition.
- These instances support RTX features and are paired with high-frequency CPUs to provide high-performance 3D graphics virtualization, making them ideal for remote graphic design, cloud gaming, and other demanding graphics processing workloads.
- Powered by Intel Ice Lake processors, these instances deliver excellent performance for 3D modeling in fields such as film and animation production, cloud gaming, and mechanical design.
Compute
- These instances are equipped with NVIDIA A10 GPUs.
  - They are based on the innovative NVIDIA Ampere architecture.
  - They support common acceleration features such as vGPU, RTX, and TensorRT.
- Processor: 2.9 GHz Intel^® Xeon^® Scalable (Ice Lake) processors with an all-core turbo frequency of 3.5 GHz.
Storage
- These are I/O optimized instances.
- Supported cloud disk types: ESSD cloud disks, ESSD AutoPL cloud disks, and ESSD Intra-city Redundant cloud disks. For more information about cloud disks, see Block storage overview.
Network
- These instances support IPv4 and IPv6. For more information about IPv6 communication, see IPv6 communication.
- Network performance scales with the instance type. Larger instance types offer better network performance.

The vgn7i-vws instance family includes the following instance types and specifications.

Instance type	vCPUs	Memory (GiB)	GPU	GPU memory	Baseline bandwidth (Gbit/s)	Forwarding rate (pps)	NIC queues	ENIs	Private IPv4 addresses	IPv6 addresses
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

Note

The GPU column in the preceding table indicates the GPU model and GPU slicing information. GPU slicing divides a physical GPU into multiple slices, allowing each instance to use one slice. For example:

In NVIDIA A10 * 1/6, NVIDIA A10 is the GPU model, and 1/6 indicates that a physical GPU is divided into six slices and the instance uses one of them.

vgn6i-vws, vGPU-accelerated instance family

Important

Alibaba Cloud has upgraded the vgn6i instance family to vgn6i-vws. The new instance family uses the latest NVIDIA GRID driver and includes a complimentary GRID vWS license. To request a free image with the GRID driver pre-installed, submit a ticket.
If you need to use a public image or a custom image that does not include the GRID driver, submit a ticket to request the GRID driver package for separate installation. Alibaba Cloud does not charge an additional license fee for the GRID driver.

Use cases:
- Real-time rendering for cloud gaming.
- Real-time rendering for augmented reality (AR) and virtual reality (VR) applications.
- AI (DL and ML) inference for elastically deployed internet services.
- Hands-on training environments for deep learning.
- Model experimentation environments for deep learning.
Compute:
- Equipped with NVIDIA T4 GPU accelerators.
- These instances use vGPUs created by GPU slicing.
  - Provides 1/4 or 1/2 the compute capability of a full NVIDIA Tesla T4 GPU.
  - Offers 4 GB or 8 GB of GPU memory.
- Features a vCPU-to-memory ratio of approximately 1:5.
- Processor: 2.5 GHz Intel ^® Xeon ^® Platinum 8163 (Skylake).
Storage:
- All instances in this family are I/O optimized.
- Supported cloud disk types: ESSDs, ESSD AutoPL cloud disks, Regional ESSDs, standard SSDs, and ultra disks. For more information about cloud disks, see Elastic Block Storage overview.
Network:
- Supports IPv4 and IPv6. For more information, see IPv6 communication.
- Network performance scales with the instance type.

The following table lists the instance types and specifications for the vgn6i-vws instance family.

Instance type	vCPUs	Memory (GiB)	GPU	GPU memory	Baseline bandwidth (Gbit/s)	Forwarding rate (pps)	NIC queues	ENIs	Private IPv4 addresses	IPv6 addresses
ecs.vgn6i-m4-vws.xlarge	4	23	NVIDIA T4 * 1/4	16 GB * 1/4	2	500,000	4/2	3	10	1
ecs.vgn6i-m8-vws.2xlarge	10	46	NVIDIA T4 * 1/2	16 GB * 1/2	4	800,000	8/2	4	10	1
ecs.vgn6i-m16-vws.5xlarge	20	92	NVIDIA T4 * 1	16 GB * 1	7.5	1,200,000	6	4	10	1

Note

The GPU column in the preceding table indicates the GPU model and GPU slicing information. GPU slicing divides a physical GPU into multiple slices, and each instance uses one slice.

In NVIDIA T4 * 1/4, NVIDIA T4 is the GPU model, and 1/4 indicates that one GPU is divided into four slices, with each instance using one.

gn9gc, GPU-accelerated compute-optimized instance family

Note

gn9gc is in invitational preview. To use gn9gc, submit a ticket.

Overview: gn9gc is Alibaba Cloud's 9th-generation cost-effective GPU cloud server instance family. It uses the latest-generation CIPU 2.0 to deliver cloud service capabilities, features high clock speed processors, and is configured with appropriate memory capacity. This instance family provides cost-effective instances for large language model (LLM) generation scenarios and video/image generation scenarios. The GPU can also directly provide graphics processing capabilities to support various rendering workloads.
Use cases:
- LLM inference: The new-generation GPU delivers compute power beyond the 8th generation with significantly improved memory bandwidth. Newly supported FP4 compute comprehensively improves inference performance and cost-effectiveness. Multi-GPU parallel inference efficiency is greatly enhanced.

Compute:

Uses the latest CIPU 2.0 cloud processor.
- The 2nd-generation CIPU provides higher cloud processing power with enhanced eRDMA, VPC, and EBS component capabilities. Supports containers (including but not limited to Docker, Clear Container, and Pouch).
Uses the new Blackwell architecture professional graphics card:
- Supports OpenGL professional-grade graphics processing.
- Supports RTX, TensorRT, and other common acceleration features, with newly upgraded FP4 support and PCIe Gen5 interconnect.

Key GPU specifications:

GPU architecture	GPU memory	Computing performance	Video encoding/decoding	Inter-GPU interconnect	Acceleration APIs
NVIDIA Blackwell	Capacity: 72 GB Bandwidth: 1,344 GB/s	TF32: 126 TFLOPS FP32: 52 TFLOPS FP16/BF16: 266 TFLOPS FP8/INT8: 530 TFLOPS FP4: 970 TFLOPS RT Core: 196 TFLOPS	3 x Video Encoder 3 x Video Decoder	PCIe interface: PCIe Gen5 x16 Bandwidth: 128 GB/s, P2P supported	DX12, OpenGL 4.6, Vulkan 1.3, CUDA 12.8, OpenCL 3.0, DirectCompute

Storage:
- I/O optimized.
- Supports the NVMe protocol. For more information, see NVMe protocol.
- Supported cloud disk types: elastic ephemeral disks, ESSDs, ESSD AutoPL disks, and regional ESSDs. For more information, see Block storage overview.
Network:
- Supports IPv4 and IPv6. For more information about IPv6, see IPv6.
- Ultra-high network performance with up to 30 million PPS (8-GPU instances).
- Supports ERI (Elastic RDMA Interface) for RDMA direct acceleration over VPC networks, with bandwidth up to 360 Gbit/s. Suitable for autonomous driving, embodied intelligence, computer vision, and traditional model training workloads.
- Note
  For more information about ERI, see Enable eRDMA on an enterprise-level instance.

The following table describes the instance types in the gn9gc instance family.

Instance type	vCPUs	Memory (GiB)	GPU memory	Baseline/burst bandwidth (Gbit/s)	Packet forwarding rate (pps)	IPv4 addresses per ENI	IPv6 addresses per ENI	NIC queues (primary/secondary)	ENIs	Max data disks	Max disk bandwidth (GB/s)
ecs.gn9gc.4xlarge	16	128	72 GB × 1	16	3.6 million	30	30	8/32	8	1	1
ecs.gn9gc.8xlarge	32	192	72 GB × 1	32	7.5 million	30	30	16/64	8	1	1
ecs.gn9gc-2x.16xlarge	64	384	72 GB × 2	65	15 million	30	30	32/64	15	2	2
ecs.gn9gc-4x.32xlarge	128	768	72 GB × 4	131	30 million	50	50	64/64	15	4	4
ecs.gn9gc-8x.64xlarge	256	1,536	72 GB × 8	204	30 million	50	50	128/64	15	6	6

Note

Images used for gn9gc instances must be in the UEFI boot mode. If you want to use a custom image, make sure that the custom image supports UEFI boot mode and that the boot mode attribute of the image is set to UEFI. For more information, see Set the boot mode of a custom image to UEFI by calling API operations.

gn8v and gn8v-tee, GPU-accelerated compute-optimized instance family

These instance families are available in select regions, including those outside the Chinese mainland. To use them, contact your Alibaba Cloud sales representative.

Introduction:
- gn8v: An 8th-generation GPU-accelerated compute-optimized instance family from Alibaba Cloud for AI model training and inference on ultra-large language models (LLMs). This family provides instance types with one, two, four, or eight GPUs for various application requirements.
- gn8v-tee: To enhance security for large model training and inference, Alibaba Cloud offers gn8v-tee, an 8th-generation instance family based on gn8v with a confidential computing feature. These instances encrypt data during GPU computation to protect your data.
Use cases:
- Cost-effective for multi-GPU parallel inference on LLMs with more than 70 billion parameters.
- Each GPU provides 39.5 TFLOPS of FP32 compute power and delivers outstanding performance for traditional AI model training and autonomous driving training workloads.
- The eight GPUs support NVLink interconnectivity and are suitable for training small- to medium-sized models.
Features:
- High-speed, large-capacity GPU memory: Each GPU is equipped with 96 GB of HBM3 GPU memory and provides up to 4 TB/s of memory bandwidth, significantly accelerating model training and inference.
- High inter-GPU bandwidth: Multiple GPUs are interconnected with NVLink at 900 GB/s. This enables much higher efficiency for multi-GPU training and inference compared to previous-generation GPU instances.
- LLM quantization: Supports FP8 compute power, which optimizes performance for large-scale parameter training and inference. This significantly improves training and inference speeds and reduces GPU memory usage.
- (For gn8v-tee instances only) High security: Supports both CPU confidential computing with Intel® Trust Domain Extensions (TDX) and GPU confidential computing with NVIDIA Confidential Computing (CC). This provides end-to-end confidential computing for the entire model inference pipeline, protecting your inference data and enterprise models during model training and inference.
Compute:
- Powered by the latest CIPU 1.0.
  - Decouples compute from storage, letting you flexibly select the storage resources you need.
  - Provides bare metal capabilities, which support peer-to-peer (P2P) communication between GPU instances, unlike traditional virtualized instances.
- Powered by 4th-generation Intel® Xeon® Scalable processors with a base frequency of up to 2.8 GHz and an all-core turbo frequency of up to 3.1 GHz.
Storage:
- I/O-optimized instance.
- These instances support the NVMe protocol. For more information, see Overview of the NVMe protocol.
- Supported cloud disk types: elastic ephemeral disk, ESSD, ESSD AutoPL disks, and Regional ESSD. For more information about cloud disks, see block storage overview.
Network:
- Supports IPv4 and IPv6. For more information about IPv6 communication, see IPv6 communication.
- These instances support jumbo frames. For more information, see Jumbo frames.
- Delivers ultra-high network performance with a packet forwarding rate of up to 30 million pps (on 8-GPU instances).
- Supports elastic RDMA interface (ERI).
- Note
  For information about how to use ERI, see Enable on enterprise-level instance.
Security: Supports the trusted computing feature (vTPM). This feature is available on gn8v instances but not on gn8v-tee instances. For more information, see Overview of trusted computing capabilities.

The following table describes the instance types in the gn8v family.

Instance type	vCPUs	Memory (GiB)	GPU memory	Network bandwidth (Gbit/s)	ENIs	Primary ENI queues	IPv4 addresses per ENI	IPv6 addresses per ENI	Max cloud disks	Baseline IOPS	Baseline bandwidth (GB/s)
ecs.gn8v.4xlarge	16	96	96 GB × 1	12	8	16	30	30	17	100,000	0.75
ecs.gn8v.6xlarge	24	128	96 GB × 1	15	8	24	30	30	17	120,000	0.937
ecs.gn8v-2x.8xlarge	32	192	96 GB × 2	20	8	32	30	30	25	200,000	1.25
ecs.gn8v-4x.8xlarge	32	384	96 GB × 4	20	8	32	30	30	25	200,000	1.25
ecs.gn8v-2x.12xlarge	48	256	96 GB × 2	25	8	48	30	30	33	300,000	1.50
ecs.gn8v-8x.16xlarge	64	768	96 GB × 8	32	8	64	30	30	33	360,000	2.5
ecs.gn8v-4x.24xlarge	96	512	96 GB × 4	50	15	64	30	30	49	500,000	3
ecs.gn8v-8x.48xlarge	192	1024	96 GB × 8	100	15	64	50	50	65	1,000,000	6

The following table describes the instance types in the gn8v-tee family.

Instance type	vCPUs	Memory (GiB)	GPU memory	Network bandwidth (Gbit/s)	ENIs	Primary ENI queues	IPv4 addresses per ENI	IPv6 addresses per ENI	Max cloud disks	Baseline IOPS	Baseline bandwidth (GB/s)
ecs.gn8v-tee.4xlarge	16	96	96 GB × 1	12	8	16	30	30	17	100,000	0.75
ecs.gn8v-tee.6xlarge	24	128	96 GB × 1	15	8	24	30	30	17	120,000	0.937
ecs.gn8v-tee-8x.16xlarge	64	768	96 GB × 8	32	8	64	30	30	33	360,000	2.5
ecs.gn8v-tee-8x.48xlarge	192	1024	96 GB × 8	100	15	64	50	50	65	1,000,000	6

Note

The gn8v-tee instance family supports only Alibaba Cloud Linux 3 images. If you use a custom image built on Alibaba Cloud Linux 3 to create an instance, ensure the kernel version is 5.10.134-18 or later.

gn8is, GPU-accelerated compute-optimized instance family

This instance family is available in select regions, including those outside the Chinese mainland. To use this instance family, contact your Alibaba Cloud sales representative.

Introduction: gn8is is Alibaba Cloud's eighth-generation GPU-accelerated compute-optimized instance family, designed for the growing demands of AI-generated content (AIGC). Powered by the latest NVIDIA L20 GPUs, this family offers instance types with one, two, four, or eight GPUs, and various CPU-to-GPU ratios to meet diverse application needs.
Features:
- Graphics processing: Powered by 4th-generation Intel® Xeon® Scalable high-frequency processors, these instances provide robust CPU compute power for 3D modeling scenarios, ensuring smoother graphics rendering and design workflows.
- Inference tasks: Equipped with the new NVIDIA L20 GPU, each with 48 GB of GPU memory, these instances accelerate inference tasks. They support the FP8 floating-point format and can be paired with Container Service for Kubernetes (ACK) to flexibly run inference for various AIGC models. They are especially suitable for inference tasks on large language models (LLMs) with fewer than 70 billion parameters.
Use cases:
- Use GRID drivers with images from Alibaba Cloud Marketplace to enable OpenGL and Direct3D capabilities. This provides workstation-grade graphics processing for workloads such as animation, film and television special effects, and rendering.
- Use the container management capabilities of Container Service for Kubernetes (ACK) for more efficient and cost-effective AIGC image generation and LLM inference.
- Other general-purpose AI applications, such as image recognition and speech recognition.

Compute:

Powered by the new NVIDIA L20 enterprise-grade GPUs.
- Supports common acceleration features such as TensorRT and the FP8 floating-point format to improve model inference performance.
- Up to 48 GB of GPU memory per GPU. With multiple GPUs, instances in this family support single-instance inference for models with 70 billion or more parameters.
- Enhanced graphics processing capabilities. After you install a GRID driver using Cloud Assistant or an image from Alibaba Cloud Marketplace, the graphics processing performance is twice that of 7th-generation platforms.

Key parameters of the NVIDIA L20 GPU:

GPU architecture	GPU memory	Compute performance	Video encoding/decoding	Inter-GPU connectivity
NVIDIA Ada Lovelace	Capacity: 48 GB Bandwidth: 864 GB/s	FP64: N/A FP32: 59.3 TFLOPS FP16/BF16: 119 TFLOPS FP8/INT8: 237 TFLOPS	3 × Video Encoders (+AV1) 3 × Video Decoders 4 × JPEG Decoders	PCIe interface: PCIe Gen4 x16 Bandwidth: 64 GB/s

Processor: Powered by the latest high-frequency Intel^® Xeon^® processors with an all-core turbo frequency of up to 3.9 GHz to handle complex 3D modeling demands.

Storage:
- All instances in this family are I/O-optimized instances.
- These instances support the NVMe protocol. For more information, see Overview of the NVMe protocol.
- Supported cloud disk types: elastic ephemeral disks, ESSDs, ESSD AutoPL disks, and Regional ESSDs. For more information about cloud disks, see Block storage overview.
Network:
- Supports IPv4 and IPv6. For more information about IPv6 communication, see IPv6 communication.
- Supports Elastic RDMA Interface (ERI).
  
  Note
  For details on using ERI, see Enable eRDMA for enterprise-level instances.
Security: These instances support the vTPM feature. For more information, see Overview of trusted computing.

The following table describes the instance types and specifications for the gn8is family.

Instance type	vCPUs	Memory (GiB)	GPU	GPU memory	Network bandwidth (Gbit/s)	ENIs	Primary ENI queues	Private IPv4 addresses	IPv6 addresses	Max cloud disks	Disk IOPS	Disk bandwidth (GB/s)
ecs.gn8is.2xlarge	8	64	L20 × 1	48 GB × 1	8	4	8	15	15	17	60,000	0.75
ecs.gn8is.4xlarge	16	128	L20 × 1	48 GB × 1	16	8	16	30	30	17	120,000	1.25
ecs.gn8is-2x.8xlarge	32	256	L20 × 2	48 GB × 2	32	8	32	30	30	33	250,000	2
ecs.gn8is-4x.16xlarge	64	512	L20 × 4	48 GB × 4	64	8	64	30	30	33	450,000	4
ecs.gn8is-8x.32xlarge	128	1024	L20 × 8	48 GB × 8	100	15	64	50	50	65	900,000	8

gn7e, GPU-accelerated compute-optimized instance family

Features of the gn7e instance family include:

Overview:
- this instance family lets you select instance types with different numbers of GPUs and CPU resources to meet your various AI business needs.
- Built on the third-generation X-Dragon architecture, gn7e instances deliver double the average network bandwidth for VPCs and cloud disks compared to the previous generation.
Use cases:
- Small- and medium-scale AI training workloads.
- High-performance computing (HPC) workloads accelerated using CUDA.
- AI inference workloads that require high GPU compute performance or large GPU memory.
- Deep learning, such as training AI algorithms for image classification, autonomous driving, and speech recognition.
- GPU-intensive scientific computing, such as computational fluid dynamics, computational finance, molecular dynamics, and environmental analysis.
Important
When you run AI training workloads with high communication loads, such as those involving Transformer models, you must enable NVLink for GPU-to-GPU communication. Otherwise, large-scale data transfers over the PCIe link may cause unexpected failures and data corruption. If you are unsure about the communication link topology for your training workload, submit a ticket for support from Alibaba Cloud technical experts.
Storage:
- All instances in this family are I/O optimized.
- Supported cloud disk types: ESSD cloud disks, ESSD AutoPL cloud disks, and ESSD Intra-city Redundant cloud disks. For more information, see Block storage overview.
Network:
- Supports IPv4 and IPv6. For more information, see IPv6 communication.
- Network performance scales with the instance type. Larger instance types offer better network performance.

The gn7e instance family includes the instance types and specifications described in the following table.

Instance type	vCPUs	Memory (GiB)	GPU memory	Baseline bandwidth (Gbit/s)	Forwarding rate (pps)	Queues	ENIs	Private IPv4 addresses	IPv6 addresses
ecs.gn7e-c16g1.4xlarge	16	125	80 GB × 1	8	3,000,000	8	8	10	1
ecs.gn7e-c16g1.8xlarge	32	250	80 GB × 2	16	6,000,000	16	8	10	1
ecs.gn7e-c16g1.16xlarge	64	500	80 GB × 4	32	12,000,000	32	8	10	1
ecs.gn7e-c16g1.32xlarge	128	1000	80 GB × 8	64	24,000,000	32	16	15	1

gn7i, GPU-accelerated compute-optimized instance family

Overview: Powered by the third-generation SHENLONG architecture, gn7i instances deliver stable and predictable high performance. They use chip-level fast path acceleration to increase storage performance, network performance, and compute stability by an order of magnitude.
Use cases:
- Equipped with high-performance CPUs, memory, and GPUs, these instances are ideal for concurrent AI inference tasks, such as image recognition, speech recognition, and behavior recognition.
- These instances support RTX features and use high-frequency CPUs to deliver high-performance 3D graphics virtualization. They are suitable for graphics-intensive workloads, such as remote graphics design and cloud gaming.
Compute:
- Equipped with NVIDIA A10 GPUs that feature:
  - The innovative NVIDIA Ampere architecture.
  - Support for common acceleration features such as RTX and TensorRT.
- Processor: 2.9 GHz Intel ^® Xeon ^® Scalable (Ice Lake) processor with an all-core turbo frequency of 3.5 GHz.
- This instance family provides up to 752 GiB of memory, a significant increase compared to the gn6i instance family.
Storage:
- All instances in this family are I/O optimized.
- Supported cloud disk types: ESSD cloud disks, ESSD AutoPL cloud disks, and ESSD Zone-redundant cloud disks. For more information, see Block storage overview.
Network:
- These instances support IPv4 and IPv6. For more information, see IPv6 communication.
- Network performance scales with the instance type. Larger instance types offer better network performance.

The gn7i instance family includes the following instance types and specifications.

Instance type	vCPUs	Memory (GiB)	GPU	GPU memory	Network bandwidth (Gbit/s)	Packet rate (PPS)	NIC queues	ENIs	Private IPv4 addresses	IPv6 addresses
ecs.gn7i-c8g1.2xlarge	8	30	NVIDIA A10 * 1	24 GB * 1	16	1,600,000	8	4	15	15
ecs.gn7i-c16g1.4xlarge	16	60	NVIDIA A10 * 1	24 GB * 1	16	3,000,000	8	8	30	30
ecs.gn7i-c32g1.8xlarge	32	188	NVIDIA A10 * 1	24 GB * 1	16	6,000,000	12	8	30	30
ecs.gn7i-c32g1.16xlarge	64	376	NVIDIA A10 * 2	24 GB * 2	32	12,000,000	16	15	30	30
ecs.gn7i-c32g1.32xlarge	128	752	NVIDIA A10 * 4	24 GB * 4	64	24,000,000	32	15	30	30
ecs.gn7i-c48g1.12xlarge	48	310	NVIDIA A10 * 1	24 GB * 1	16	9,000,000	16	8	30	30
ecs.gn7i-c56g1.14xlarge	56	346	NVIDIA A10 * 1	24 GB * 1	16	10,000,000	16	8	30	30
ecs.gn7i-2x.8xlarge	32	128	NVIDIA A10 * 2	24 GB * 2	16	6,000,000	16	8	30	30
ecs.gn7i-4x.8xlarge	32	128	NVIDIA A10 * 4	24 GB * 4	32	6,000,000	16	8	30	30
ecs.gn7i-4x.16xlarge	64	256	NVIDIA A10 * 4	24 GB * 4	64	12,000,000	32	8	30	30
ecs.gn7i-8x.32xlarge	128	512	NVIDIA A10 * 8	24 GB * 8	64	24,000,000	32	16	30	30
ecs.gn7i-8x.16xlarge	64	256	NVIDIA A10 * 8	24 GB * 8	32	12,000,000	32	8	30	30

Important

You can change instances of the types ecs.gn7i-2x.8xlarge, ecs.gn7i-4x.8xlarge, ecs.gn7i-4x.16xlarge, ecs.gn7i-8x.32xlarge, and ecs.gn7i-8x.16xlarge to ecs.gn7i-c8g1.2xlarge or ecs.gn7i-c16g1.4xlarge. However, you cannot change them to other instance types such as ecs.gn7i-c32g1.8xlarge.

gn7s, GPU-accelerated compute-optimized instance family

To use the gn7s instance family, submit a ticket.

Introduction:
- This instance family is powered by the latest Intel Ice Lake processors and NVIDIA A30 GPUs based on the NVIDIA Ampere architecture. This family offers various instance types with different GPU and CPU configurations to meet your specific AI needs.
- Built on Alibaba Cloud's third-generation SHENLONG architecture, gn7s instances deliver twice the average network bandwidth for VPCs and cloud disks as the previous generation.
Use cases: Featuring high-performance CPUs, memory, and GPUs, these instances are ideal for concurrent AI inference workloads, such as image recognition, speech recognition, and behavior identification.
Compute:
- Features NVIDIA A30 GPUs, which include:
  - The innovative NVIDIA Ampere architecture.
  - Support for the Multi-Instance GPU (MIG) feature and acceleration based on second-generation Tensor Cores for a wide range of workloads.
- Processor: 2.9 GHz Intel ^® Xeon ^® Scalable (Ice Lake) processor with an all-core turbo frequency of 3.5 GHz.
- Offers significantly more memory than the previous-generation instance family.
Storage:
- All instances in this family are I/O optimized.
- Supported cloud disk types: ESSD, ESSD AutoPL, and Zone-redundant ESSD. For more information, see Block storage overview.
Network:
- Supports IPv4 and IPv6. For more information, see IPv6 communication.
- Network performance scales with the instance type.

The gn7s instance family includes the following instance types and specifications:

Instance type	vCPUs	Memory (GiB)	GPUs	GPU memory	Network bandwidth (Gbit/s)	Packet rate (pps)	Private IPv4s per ENI	IPv6s per ENI	Multi-queue	ENIs
ecs.gn7s-c8g1.2xlarge	8	60	NVIDIA A30 * 1	24GB * 1	16	1,600,000	5	1	8	4
ecs.gn7s-c16g1.4xlarge	16	120	NVIDIA A30 * 1	24GB * 1	16	3,000,000	5	1	8	8
ecs.gn7s-c32g1.8xlarge	32	250	NVIDIA A30 * 1	24GB * 1	16	6,000,000	5	1	12	8
ecs.gn7s-c32g1.16xlarge	64	500	NVIDIA A30 * 2	24GB * 2	32	12,000,000	5	1	16	15
ecs.gn7s-c32g1.32xlarge	128	1000	NVIDIA A30 * 4	24GB * 4	64	24,000,000	10	1	32	15
ecs.gn7s-c48g1.12xlarge	48	380	NVIDIA A30 * 1	24GB * 1	16	9,000,000	8	1	16	8
ecs.gn7s-c56g1.14xlarge	56	440	NVIDIA A30 * 1	24GB * 1	16	10,000,000	8	1	16	8

gn7, GPU-accelerated compute-optimized instance family

Scenarios:
- Deep learning, such as training AI algorithms used in image classification, autonomous driving, and speech recognition.
- GPU-intensive scientific computing, such as computational fluid dynamics, computational finance, molecular dynamics, and environmental analysis.

Storage:
- Instances are I/O optimized.
- Supports ESSD cloud disks, ESSD AutoPL cloud disks, and ESSD Zone-redundant cloud disks. For more information, see Block storage overview.
Network:
- Supports IPv4 and IPv6. For more information about IPv6, see IPv6 communication.
- Network performance scales with the instance type.

The following table describes the instance types and specifications of the gn7 instance family.

Instance type	vCPUs	Memory (GiB)	GPU memory	Network bandwidth (Gbit/s)	Packet rate (pps)	NIC queues	ENIs	Private IPv4 addresses	IPv6 addresses
ecs.gn7-c12g1.3xlarge	12	94	40 GB × 1	4	2,500,000	4	8	10	1
ecs.gn7-c13g1.13xlarge	52	378	40 GB × 4	16	9,000,000	16	8	30	30
ecs.gn7-c13g1.26xlarge	104	756	40 GB × 8	30	18,000,000	16	15	10	1

gn6i, GPU-accelerated compute-optimized instance family

Use cases:
- AI (deep learning and machine learning) inference for applications such as computer vision, speech recognition, speech synthesis, natural language processing (NLP), machine translation, and recommendation systems.
- Real-time rendering for cloud gaming.
- Real-time, cloud-based rendering for augmented reality (AR) and virtual reality (VR).
- Graphics-heavy computing or graphics workstations.
- GPU-accelerated databases.
- High-performance computing (HPC).
Compute:
- Equipped with NVIDIA T4 GPU accelerators, featuring:
  - The innovative NVIDIA Turing architecture.
  - 16 GB of memory per GPU with a memory bandwidth of 320 GB/s.
  - 2,560 CUDA cores per GPU.
  - Up to 320 Turing Tensor Cores per GPU.
  - Mixed-precision Tensor Cores that support 65 TFLOPS of FP16, 130 TOPS of INT8, and 260 TOPS of INT4.
- vCPU-to-memory ratio of approximately 1:4.
- Processor: 2.5 GHz Intel^® Xeon^® Platinum 8163 (Skylake).
Storage:
- I/O-optimized instances.
- Supported disk types: ESSD, ESSD AutoPL disks, SSD disk, and ultra disk. For more information, see Block storage overview.
Network:
- Supports IPv4 and IPv6. For details, see IPv6 communication.
- Network performance scales with the instance type.

The gn6i instance family includes the following instance types.

Instance type	vCPUs	Memory (GiB)	GPUs	GPU memory	Network bandwidth (Gbit/s)	Packet rate (pps)	Disk IOPS	Multi-queue	ENIs	Private IPv4 addresses	IPv6 addresses
ecs.gn6i-c4g1.xlarge	4	15	NVIDIA T4 × 1	16 GB × 1	4	2,500,000	N/A	2	2	10	1
ecs.gn6i-c8g1.2xlarge	8	31	NVIDIA T4 × 1	16 GB × 1	5	2,500,000	N/A	2	2	10	1
ecs.gn6i-c16g1.4xlarge	16	62	NVIDIA T4 × 1	16 GB × 1	6	2,500,000	N/A	4	3	10	1
ecs.gn6i-c24g1.6xlarge	24	93	NVIDIA T4 × 1	16 GB × 1	7.5	2,500,000	N/A	6	4	10	1
ecs.gn6i-c40g1.10xlarge	40	155	NVIDIA T4 × 1	16 GB × 1	10	2,500,000	N/A	16	10	10	1
ecs.gn6i-c24g1.12xlarge	48	186	NVIDIA T4 × 2	16 GB × 2	15	4,500,000	N/A	12	6	10	1
ecs.gn6i-c24g1.24xlarge	96	372	NVIDIA T4 × 4	16 GB × 4	30	4,500,000	250,000	24	8	10	1

gn6e, GPU-accelerated compute-optimized instance family

Use cases:
- Deep learning applications, such as training and inference for AI algorithms for image classification, autonomous driving, and speech recognition.
- Scientific computing, such as computational fluid dynamics, computational finance, molecular dynamics, and environmental analysis.
Compute:
- Features NVIDIA V100 (32 GB NVLink) GPU cards.
- GPU accelerator: V100 (SXM2 package).
  - Innovative NVIDIA Volta architecture.
  - 32 GB of HBM2 memory per GPU with a GPU memory bandwidth of 900 GB/s.
  - 5,120 CUDA Cores per GPU.
  - 640 Tensor Cores per GPU.
  - Each GPU supports six bidirectional NVLink connections, each providing 25 Gbit/s of bandwidth in each direction for a total of 300 Gbit/s.
- Features a vCPU-to-memory ratio of approximately 1:8.
- Processor: 2.5 GHz Intel ^® Xeon ^® Platinum 8163 (Skylake).
Storage:
- I/O optimized instance.
- Supported cloud disk types: ESSDs, ESSD AutoPL disks, Regional ESSDs, standard SSDs, and ultra disks. For more information, see Elastic Block Storage.
Network:
- Supports both IPv4 and IPv6. For more information about IPv6 communication, see IPv6 communication.
- Network performance scales with the instance type.

gn6e includes the instance types and specifications listed in the table below.

Instance type	vCPU	Memory (GiB)	GPU	GPU memory	Baseline bandwidth (Gbit/s)	Packet rate (PPS)	NIC queues	ENI	Private IPv4 addresses	IPv6 addresses
ecs.gn6e-c12g1.3xlarge	12	92	1 × NVIDIA V100	1 × 32 GB	5	800,000	8	6	10	1
ecs.gn6e-c12g1.6xlarge	24	184	2 × NVIDIA V100	2 × 32 GB	8	1,200,000	8	8	20	1
ecs.gn6e-c12g1.12xlarge	48	368	4 × NVIDIA V100	4 × 32 GB	16	2,400,000	8	8	20	1
ecs.gn6e-c12g1.24xlarge	96	736	8 × NVIDIA V100	8 × 32 GB	32	4,500,000	16	8	20	1

gn6v, GPU-accelerated compute-optimized instance family

Use cases:
- Deep learning applications, such as training and inference for AI algorithms in image classification, autonomous driving, and speech recognition.
- Scientific computing, such as computational fluid dynamics, computational finance, molecular dynamics, and environmental analysis.
Compute:
- Equipped with NVIDIA V100 GPUs.
- GPU accelerator: V100 (SXM2 package).
  - Innovative NVIDIA Volta architecture.
  - 16 GB of HBM2 GPU memory per GPU with 900 GB/s of memory bandwidth.
  - 5,120 CUDA Cores per GPU.
  - 640 Tensor Cores per GPU.
  - Up to six NVLink bidirectional connections per GPU. Each connection provides a bandwidth of 25 Gbit/s in each direction, for a total bandwidth of 300 Gbit/s.
- Features a vCPU-to-memory ratio of approximately 1:4.
- Processor: 2.5 GHz Intel^® Xeon^® Platinum 8163 (Skylake).
Storage:
- All instances in this family are I/O optimized.
- Supported disk types: ESSD, ESSD AutoPL, SSD Cloud Disk, and Ultra Disk. For more information, see Block storage overview.
Network:
- Supports IPv4 and IPv6. For more information, see IPv6 communication.
- Network performance scales with the instance type.

The gn6v instance family includes the instance types and specifications listed below.

Instance type	vCPU	Memory (GiB)	GPU	GPU memory	Network bandwidth (Gbit/s)	Packet rate (pps)	Disk baseline IOPS	Multi-queue	ENI	Private IPv4 addresses	IPv6 addresses
ecs.gn6v-c8g1.2xlarge	8	32	1 × NVIDIA V100	1 × 16 GB	2.5	800,000	N/A	4	4	10	1
ecs.gn6v-c8g1.4xlarge	16	64	2 × NVIDIA V100	2 × 16 GB	5	1,000,000	N/A	4	8	20	1
ecs.gn6v-c8g1.8xlarge	32	128	4 × NVIDIA V100	4 × 16 GB	10	2,000,000	N/A	8	8	20	1
ecs.gn6v-c8g1.16xlarge	64	256	8 × NVIDIA V100	8 × 16 GB	20	2,500,000	N/A	16	8	20	1
ecs.gn6v-c10g1.20xlarge	82	336	8 × NVIDIA V100	8 × 16 GB	35	4,500,000	250,000	16	8	20	1

GPU virtualization	GPU compute	Not recommended
sgn8ia sgn7i-vws (shared CPU) vgn7i-vws vgn6i-vws	gn9gc, GPU-accelerated compute-optimized instance family gn8v/gn8v-tee gn8is gn7e gn7i gn7 gn6i gn6e gn6v	gn7s, GPU-accelerated compute-optimized instance family