Overview

The instance types listed in this document will soon stop being sold in all regions. Please refer to the official notification for the specific stop-sale date. Instance types pending discontinuation are currently available to allowlist users. Existing users can still create instances of these types, while new users are advised to purchase from the following recommended instance families.

To be discontinued soon	Recommended use
GPU Accelerated g1vc	GPU Accelerated g1ve
GPU Accelerated g1v	GPU Accelerated g1ve
GPU Accelerated g1t	GPU Accelerated g1te
HPC-optimized hpcg1v	HPC-optimized hpcg1ve

Models pending discontinuation

GPU Accelerated g1vc

Features

type	Performance
Reserved Instances	Equipped with an Intel® Xeon® Gold processor (Skylake) with a processor base frequency of 2.1 GHz and an all-core turbo frequency of 3.7 GHz.
		The processor-to-memory ratio is approximately 1 to 4.
		Supports up to 52 vCPUs and 218 GiB
		GPU: NVIDIA Tesla V100 (32 GB video memory per card), up to 8 GPUs can be mounted to a single instance.
Storage	Enhanced SSD Volume
Storage		Storage I/O performance is related to computing specifications; higher specifications deliver better performance.
Network	Maximum network bandwidth: 24 Gbit/s
		Maximum throughput: 4.5 million PPS
		Network performance is related to computing specifications; the higher the specifications, the better the performance.

Application scenarios

Machine learning training scenarios, such as image classification, autonomous driving, speech recognition, and other applications involving the training of artificial intelligence algorithms.

High-performance scientific computing and simulation, for example, computational fluid dynamics, computational finance, molecular dynamics, environmental analysis, and so on.

Specifications list

Instance type	vCPU	Memory (GiB)	GPU	GPU memory (GB)	Outbound and inbound network bandwidth capability (Gbit/s)	Maximum throughput (inbound and outbound, 10,000 PPS)	Number of connections (×10,000)	Network interface multiqueue	elastic network interface (including a primary network interface)	Private IP of a single network interface	Volume IOPS (×10,000)	Volume bandwidth (Gbit/s)	Number of volumes (including one system volume)	Disk queue count
ecs.g1vc.xlarge	6	26	V100*1	32GB * 1	3	50	20	4	4	10	4	3	15	2
ecs.g1vc.3xlarge	12	52	V100*2	32GB * 2	6	100	60	8	4	20	4	3	15	2
ecs.g1vc.6xlarge	24	104	V100*4	32GB * 4	12	180	120	16	8	20	8	5	15	4
ecs.g1vc.13xlarge	52	218	V100*8	32GB * 8	24	450	300	32	15	20	20	12	15	4

GPU Accelerated g1v

Features

type	Performance
Reserved Instances	Equipped with 2nd generation Intel® Xeon® Scalable processors (Cascade Lake), with a processor base frequency of 2.4 GHz and an all-core turbo frequency of 3.1 GHz.
		The processor-to-memory ratio is 1:4.
		Supports up to 88 vCPUs and 352 GiB
		GPU: NVIDIA Tesla V100 (32 GB video memory per card); a single instance supports mounting up to 8 GPUs.
Storage	PTSSD Volume
Storage		Storage I/O performance is related to computing specifications; the higher the specifications, the better the performance.
Network	Maximum network bandwidth: 25 Gbit/s
		Maximum throughput: 3.5 million PPS
		Network performance depends on computing specifications; the higher the specifications, the better the performance.

Application scenarios

Deep learning is used in training applications of artificial intelligence algorithms, such as image classification, autonomous driving, and speech recognition.

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

List of specifications

Instance type	vCPU	Memory (GiB)	GPU	GPU memory (GB)	Network bandwidth capability (outbound + inbound) (Gbit/s)	Maximum throughput (inbound + outbound, 10,000 PPS)	Number of connections (in 10,000s)	Network interface multiqueue	Elastic network interfaces (including one primary network interface)	Private IP address for a single network interface	Volume IOPS (×10,000)	Volume bandwidth (Gbit/s)	Number of volumes (including one system volume)
ecs.g1v.2xlarge	8	32	V100 * 1	32GB * 1	2	30	25	8	4	10	2.5	1.5	15
ecs.g1v.8xlarge	32	128	V100 * 4	32GB * 4	8.5	150	60	16	8	20	6	7.5	15
ecs.g1v.22xlarge	88	352	V100 * 8	32GB * 8	25	350	180	32	15	20	20	10	15

GPU Accelerated g1t

Features

type	Performance
Reserved Instances	Equipped with 2nd Generation Intel® Xeon® Scalable processors (Cascade Lake), with a processor base frequency of 2.4 GHz and an all-core turbo frequency of 3.1 GHz.
		The processor-to-memory ratio is 1:4.
		Supports up to 88 vCPUs and 226 GiB
		GPU: NVIDIA Tesla T4 (16 GB video memory per card), each instance supports mounting up to 8 GPUs.
Storage	PTSSD Volume
Storage		Storage I/O performance is related to computing specifications; higher specifications provide better performance.
Network	Maximum network bandwidth: 25 Gbit/s
		No limit on network packet size
		Network performance is related to computing specifications; the higher the specifications, the better the performance.

Application scenarios

It is suitable for processing AI computer vision, speech recognition, speech synthesis, natural language processing, machine translation, and recommendation systems.

Video rendering and graphics workstations.

Specifications

Instance type	vCPU	Memory (GiB)	GPU	GPU memory (GB)	Outbound + inbound network bandwidth capability (Gbit/s)	Maximum throughput (outbound + inbound, 10,000 PPS)	Number of connections (×10,000)	Network interface multiqueue support	Elastic network interface (including a primary network interface)	Private IP of a single network interface	Volume IOPS (×10,000)	Volume bandwidth (Gbit/s)	Number of volumes (including one system volume)
ecs.g1t.xlarge	4	16	T4 * 1	16GB * 1	1	No limit	25	8	2	10	2	1.2	15
ecs.g1t.2xlarge	8	28	T4 * 1	16GB * 1	2	No limit	25	8	2	10	2.5	1.5	15
ecs.g1t.4xlarge	16	56	T4 * 2	16GB * 2	4	No limit	30	31	2	20	4	2.5	15
ecs.g1t.8xlarge	32	112	T4 * 4	16GB * 4	8.5	Unlimited	60	31	2	20	6	7.5	15
ecs.g1t.22xlarge	88	226	T4 * 8	16GB * 8	25	No limit	180	31	2	20	20	10	15

HPC-optimized hpcg1v

Features

type	Performance
Reserved Instances	Featuring 2nd Gen Intel® Xeon® Scalable processors (Cascade Lake), with a processor base frequency of 2.4 GHz and an all-core turbo frequency of 3.1 GHz.
		Supports 88 vCPUs and 350 GiB
		GPU: NVIDIA V100 (32 GB video memory per card); a single instance supports mounting 8 GPUs
Storage	PTSSD Volume
Network	Maximum network bandwidth: 25 Gbit/s
		Maximum throughput: 3.5 million PPS
		Up to 100 × 1 Gbps RDMA interconnect bandwidth is supported between instances

Application scenarios

Large-scale machine learning training scenarios.

High-performance scientific computing and simulation.

Specification list

Instance type

vCPU

Memory (GiB)

GPU

GPU memory (GB)

RDMA network (Gbps)

Outbound and inbound network bandwidth capacity (Gbit/s)

Maximum throughput (outbound + inbound) (kpps)

Number of connections (×10,000)

Multiqueue network interface

An elastic network interface (including a primary network interface)

Private IP for a single network interface

Volume IOPS (×10,000)

Volume bandwidth (Gbit/s)

Number of volumes (including one system volume)

ecs.hpcg1v.22xlarge

88

350

V100 * 8

32GB * 8

100 * 1

25

350

200

32

15

20

10

15

Impact of sales suspension

After sales are discontinued, you will not be able to purchase instances from the GPU Accelerated g1v, GPU Accelerated g1t, or HPC-optimized hpcg1v instance families in any region or availability zone.

For ECS instances in the GPU Accelerated g1v, GPU Accelerated g1t, and HPC-optimized hpcg1v instance families that have already been purchased, you can still purchase PTSSD Volumes and mount them for use. PTSSD Volumes that have already been mounted can be scaled up as usual. However, usage is affected in the following scenarios:

If a pay-as-you-go instance uses the Economical Mode feature, it may not be able to start if resources are sold out.

Subscription instances can be renewed for continued use as long as they are renewed on time. If a resource is released due to overdue renewal, it cannot be recreated or restored.

Pay-as-you-go and subscription instances can no longer perform modify instance type operations (including scaling up or down).

Elastic Compute Service