Elastic bare metal server specifications

更新时间:
复制 MD 格式

ECS Bare Metal Instances let you deploy traditional, non-virtualized applications to the cloud. They seamlessly integrate with other Alibaba Cloud products, such as storage, networking, and database, so you can build solutions for various business scenarios. This topic describes the advantages and features of ECS Bare Metal Instance families and lists the available instance types.

Note

ECS Bare Metal Instances

ECS Bare Metal Instance is a new category of compute server built on Alibaba Cloud's proprietary, next-generation virtualization technology. It combines the elasticity of a virtual machine with the performance and features of a physical machine. This next-generation technology retains the elasticity of a standard ECS instance while delivering the full performance and features of a physical machine. It also supports nested virtualization.

Note

Nested virtualization is enabled by default on ECS Bare Metal Instances.

ECS Bare Metal Instances combine the advantages of physical machines and ECS instances to deliver powerful and stable computing performance. Built on Alibaba Cloud's proprietary Virtualization 2.0 technology, your applications can directly access the processor and memory of an ECS Bare Metal Instance with no virtualization overhead. These instances provide full processor features, such as Intel VT-x, and the same resource isolation as a physical machine, making them ideal for migrating applications from non-virtualized environments to the cloud.

An ECS Bare Metal Instance integrates features from both physical machines and virtual machines. It uses proprietary Alibaba Cloud technologies, including custom chips, a self-developed hypervisor, and a redefined server hardware architecture. It also seamlessly integrates with other Alibaba Cloud products, such as storage, networking, and databases. It is also compatible with standard ECS instance images, providing the flexibility to build solutions that meet your business needs.

Note the following when using ECS Bare Metal Instances:

  • You cannot change instance specifications.

  • ECS Bare Metal Instances support failover in the event of a hardware failure. Your data is retained on Cloud Disks.

  • Basic CPU monitoring metrics for EBM instances are not available through standard monitoring. To obtain this data, you must install the CloudMonitor agent. For more information, see Install CloudMonitor agent.

Benefits

Elastic Bare Metal Server provides key technical advantages.

  • Dedicated computing resources

    Elastic Bare Metal Server provides the performance and isolation of a physical machine. You get dedicated computing resources with no virtualization overhead or feature loss. You can choose from instances with various CPU core counts, including 80, 96, 104, 128, and 192 cores.

  • Confidential computing

    In addition to physical isolation, Elastic Bare Metal Server enhances data security by using a trusted execution environment (Intel® SGX). This environment ensures that your data is processed only within a secure and trusted context. This hardware-level security acts as a vault for your cloud data, giving you full control over data encryption and key protection. For more information, see Install SGX.

  • Hybrid and private cloud compatibility

    Elastic Bare Metal Server meets the demands of high-performance computing and helps you build modern hybrid clouds. It combines the flexibility of a virtual machine with the performance and features of a physical machine, enabling nested virtualization. You can seamlessly migrate your on-premises private clouds to Alibaba Cloud without the typical performance overhead of nested virtualization.

  • Support for heterogeneous instruction set architectures

    Powered by Alibaba Cloud's proprietary Virtualization 2.0, Elastic Bare Metal Server supports various instruction set architectures, such as ARM.

Instance type comparison

Elastic Bare Metal Server delivers significantly higher performance than a physical machine with the same specifications. During the Double 11 sales event, the service provided millions of vCPUs to handle the traffic peak.

The following table compares Elastic Bare Metal Server with physical machines and virtual machines. In this comparison, 'Y' indicates the feature is supported, 'N' indicates it is not, and 'N/A' means data is unavailable.

Feature category

Feature

Elastic Bare Metal Server

Physical machine

Virtual machine

O&M automation

Provisioning in minutes

Y

N

Y

compute

No performance loss

Y

Y

N

No feature loss

Y

Y

N

No resource contention

Y

Y

N

storage

Fully compatible with ECS cloud disks

Y

N

Y

Boot from a cloud disk (system disk)

Y

N

Y

Fast system disk reset

Y

N

Y

Use ECS images

Y

N

Y

Cold migration between physical machines and virtual machines

Y

N

Y

OS installation not required

Y

N

Y

Uses enhanced cloud disk data protection instead of local RAID

Y

N

Y

network

Fully compatible with VPC

Y

N

Y

Fully compatible with classic network

Y

N

Y

Seamless VPC communication between physical and virtual machine clusters

Y

N

Y

management and control

Fully compatible with the ECS management and control system

Y

N

Y

Matches the user experience of a virtual machine, including VNC

Y

N

Y

Out-of-band network security

Y

N

N/A

General-purpose instance family ebmg9ae

  • Instance family overview: This instance family is built on Alibaba Cloud's new CIPU architecture and features the latest AMD EPYC Turin processors, delivering stable compute performance, a powerful I/O engine, and chip-level security hardening.

  • Use cases:

    • Workloads that require direct access to physical resources or involve hardware-specific licensing.

    • Running third-party hypervisors to build hybrid cloud and multi-cloud deployments.

    • Containers, including but not limited to Docker, Clear Containers, and Pouch.

    • Workloads with high packet rates, such as live video commenting and telecom service forwarding.

    • Enterprise applications of various types and sizes.

    • Websites and application servers.

    • Game servers.

    • Small and medium-sized database systems, caches, and search clusters.

    • Data analytics and computing.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: AMD EPYC Turin processors with a turbo frequency of up to 3.7 GHz. Instances in this family use physical cores, ensuring stable compute performance.

  • Storage:

  • Network:

ebmg9ae includes the instance specifications and metric data listed in the following table:

Instance type

vCPUs

Memory (GiB)

Network bandwidth (Gbit/s)

Packet rate (PPS)

Connections

Queues

ENIs

IPv4 addresses per ENI

IPv6 addresses per ENI

Baseline disk IOPS

Disk bandwidth (Gbit/s)

ecs.ebmg9ae.48xlarge

192

768

100/None

30,000,000

6,000,000

64

38

50

50

600,000

50/None

General-purpose ebmg9a instance family

  • Instance family description: Built on Alibaba Cloud's new CIPU architecture and powered by the latest AMD EPYC Turin processors, this instance family delivers stable compute performance, a powerful I/O engine, and chip-level security hardening.

  • Scenarios:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Compatibility with third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Containers, and Pouch.

    • High network packet processing scenarios, such as live video comments and telecommunication service forwarding.

    • Various enterprise-grade applications.

    • Web and application servers.

    • Game servers.

    • Small and medium-sized database systems, caches, and search clusters.

    • Data analytics and computing.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: AMD EPYC Turin processors with a maximum turbo frequency of 4.1 GHz provide stable compute performance.

  • Storage:

  • Network:

The following table lists the specifications for the ebmg9a instance type family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Multi-queue

ENIs

Private IPv4 per ENI

IPv6 per ENI

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmg9a.64xlarge

256

1152

100/None

30,000,000

6,000,000

64

38

50

50

500,000

64/None

General-purpose elastic bare metal server instance family ebmg9i

  • Instance family overview: Built on the new Alibaba Cloud CIPU architecture and Intel® Xeon® 6 processors with P-cores (Performance-cores), this instance family provides stable computing power, an enhanced I/O engine, and chip-level security hardening.

  • Use cases:

    • Workloads that require direct access to physical resources or have licensing requirements tied to specific hardware.

    • Support for third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers, including Docker, Clear Containers, and Pouch.

    • High-packet-rate applications, such as live video streaming and telecom service forwarding.

    • Enterprise-grade applications of all types and sizes.

    • Websites and application servers.

    • Game servers.

    • Small and medium-sized database systems, caches, and search clusters.

    • Data analytics and computing.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: Intel® Xeon® Granite Rapids with a base frequency of 3.2 GHz and an all-core turbo frequency of 3.6 GHz.

  • Storage:

    • These are I/O optimized instances.

    • These instances support the NVMe protocol. For more information, see Overview of the NVMe protocol.

    • Supported cloud disk types: ESSD cloud disk, ESSD AutoPL cloud disk, and ESSD zone-redundant cloud disk. For more information, see Block Storage Overview.

  • Network:

The following table lists the instance types and specifications for the ebmg9i instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Multi-queue

Elastic network interfaces

Private IPv4 count

IPv6 count

Cloud disk IOPS

Cloud disk bandwidth

ecs.ebmg9i.48xlarge

192

768

64/None

20,000,000

6,000,000

64 (primary network interface)/16 (secondary network interface)

32

50

50

500,000/800,000

32/None

General-purpose ECS bare metal instance family ebmg8a

  • Instance family overview: Powered by Alibaba Cloud's new CIPU architecture and AMD EPYC Genoa processors, this instance family delivers stable computing power, a more powerful I/O engine, and chip-level security hardening.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatibility with third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Workloads that require high network packet processing rates, such as live video commenting and telecommunication service forwarding.

    • Enterprise-grade applications of all types and sizes.

    • Websites and application servers.

    • Game servers.

    • Small and medium-sized database systems, caches, and search clusters.

    • Data analytics and computing.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: AMD EPYC Genoa processors with a turbo frequency of up to 3.7 GHz, delivering stable computing performance.

  • Storage:

  • Network:

ebmg8a includes the instance specifications and metric data shown in the following table:

Instance type

vCPUs

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Multi-queue

ENIs

Private IPv4 addresses

IPv6 addresses

Cloud disk IOPS

Cloud disk throughput (Gbit/s)

ecs.ebmg8a.48xlarge

192

768

64/None

18,000,000

3,000,000

64

38

30

30

500,000/None

32/None

General-purpose elastic bare metal instance family ebmg8y

  • Instance family overview: Built on the Alibaba Cloud CIPU architecture, this instance family provides stable computing power, an enhanced I/O engine, and dedicated hardware resources with physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or depend on specific hardware for licensing.

    • Supports third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Containers, and Pouch.

    • Video on demand and live streaming.

    • Enterprise applications of all types and sizes.

    • Websites and application servers.

    • Data analytics.

    • High-performance scientific and engineering applications.

  • Compute:

    • Processor: Alibaba Cloud's Yitian 710 CPU, based on the ARM architecture, with a clock speed of at least 2.75 GHz. Hyper-threading is disabled to ensure stable compute performance.

  • Storage:

  • Network:

The following table describes the instance types and specifications of the ebmg8y instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (baseline/burst)

Network PPS

Connections

Multi-queue

ENIs

Private IPv4 per ENI

IPv6 per ENI

Disk IOPS (baseline/burst)

Disk bandwidth (baseline/burst)

ecs.ebmg8y.32xlarge

128

512

64/None

20,000,000

3,000,000

64 (primary network interface)/32 (secondary network interface)

38

30

30

500,000/None

32/None

General-purpose ECS Bare Metal Instance family ebmg8i

  • Instance family overview: Powered by the new Alibaba Cloud CIPU architecture, this instance family provides stable computing power, a more powerful I/O engine, dedicated hardware resources, and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Third-party hypervisor compatibility for hybrid cloud and multi-cloud deployment.

    • Containers, including but not limited to Docker, Clear Container, and Pouch.

    • High network packet processing, such as live video comments and telecom service forwarding.

    • Enterprise applications of all types and sizes.

    • Website and application servers.

    • Game servers.

    • Small- and medium-sized database systems, caches, and search clusters.

    • Data analysis and computing.

    • High-performance scientific and engineering applications.

  • Compute:

    • Processor: Powered by Intel®Xeon®Emerald Rapids or Intel®Xeon®Sapphire Rapids processors with a base frequency of at least 2.7 GHz and an all-core turbo frequency of 3.2 GHz, ensuring stable compute performance.

      Note

      When you purchase this instance, the system randomly assigns one of the two processor types. You cannot select a specific processor.

    • These instances support hyper-threading. Hyper-threading is enabled by default. For more information, see Change CPU options.

    • For information about operating system compatibility, see Operating system compatibility with Intel instance types.

  • Storage:

  • Network:

The following table describes the instance types and specifications of the ebmg8i instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (baseline/burst) (Gbit/s)

Network PPS

Connections

Multi-queue

ENIs

Private IPv4 addresses

IPv6 addresses

Cloud disk IOPS (baseline/burst)

Cloud disk bandwidth (baseline/burst) (Gbit/s)

ecs.ebmg8i.48xlarge

192

1024

100/None

30,000,000

4,000,000

64 (primary network interface)/16 (secondary network interface)

72

30

30

1,000,000/None

48/None

Storage-enhanced elastic bare metal instance family ebmg7se

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, these instances feature chip-level acceleration for stable, predictable, and ultra-high performance across compute, storage, and networking.

    • These instances provide dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Workloads that require high availability and the multi-attach feature.

    • I/O-intensive applications, such as medium-to-large OLTP core databases and medium-to-large NoSQL databases.

    • Search and real-time log analytics.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: Intel® Xeon® Platinum 8369B (Ice Lake) with a 2.9 GHz base frequency and a 3.5 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers network performance of up to 12 million PPS.

The following table lists the specifications for the ebmg7se instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

IPv4 addresses per ENI

IPv6 addresses per ENI

Max attached disks

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmg7se.32xlarge

128

512

32

12,000,000

2,400,000

16

30

1

32

1,000,000

64

General-purpose ECS Bare Metal instance family ebmg7

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • It provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses

    • Compatibility with third-party hypervisors to support hybrid cloud and multi-cloud deployments

    • Containers (such as Docker, Clear Container, and Pouch)

    • High packet forwarding rate scenarios, such as live video comments and telecommunication service forwarding

    • Enterprise applications of various types and sizes

    • Websites and application servers

    • Game servers

    • Small and medium-sized database systems, caches, and search clusters

    • Data analytics and computing

    • High-performance scientific and engineering applications

  • Compute:

    • vCPU-to-memory ratio of 1:4

    • Processor: Intel® Xeon® Platinum 8369B (Ice Lake) with a 2.9 GHz base frequency and a 3.5 GHz all-core turbo frequency

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For 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 24,000,000 PPS.

The following table describes the instance types and specifications of the ebmg7 instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Packet forwarding rate (PPS)

Connections

ENIs

Private IPv4 addresses

IPv6 addresses

Baseline disk IOPS

Baseline disk bandwidth (Gbit/s)

ecs.ebmg7.32xlarge

128

512

64

24,000,000

2,400,000

32

20

20

600,000

32

General-purpose ebmg7a elastic bare metal instance family

  • Instance family overview:

    • Built on the third-generation X-Dragon Architecture, this instance family uses chip-level fast-path acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • Provides dedicated hardware resources and physical isolation.

  • Scenarios:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Compute clusters and in-memory data processing.

    • Video encoding, decoding, and rendering.

    • Data analytics and compute.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: AMD EPYC™ MILAN processor with a 2.55 GHz base frequency and a single-core turbo frequency of up to 3.5 GHz, which provides stable compute performance.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers network performance of 24 million PPS.

The following table lists the instance types and performance metrics for the ebmg7a instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

Queues

ENIs

Private IPv4 addresses

IPv6 addresses

Baseline disk IOPS

Baseline disk bandwidth (Gbit/s)

ecs.ebmg7a.64xlarge

256

1024

64

24,000,000

4,000,000

32

31

15

1

600,000

32

Note
  • This instance type requires images that use the UEFI boot mode. If you use a custom image, ensure that its boot mode is set to UEFI. For more information, see Instance boot modes.

  • The kernels of Ubuntu 18 and Debian 9 do not support the AMD EPYCTM MILAN processor. Do not use images of these operating systems to create instances of this instance type, as the instances will fail to start.

General-purpose ebmg6a instance family

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level fast-path acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatibility with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Video encoding and decoding, and rendering.

    • Compute clusters and memory-intensive data processing.

    • Data analytics.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: AMD EPYCROME processor with a base frequency of 2.6 GHz and a turbo frequency of up to 3.3 GHz.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Maximum packet forwarding rate of 24,000,000 PPS.

The following table lists the instance types and metrics of the ebmg6a instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Queues

ENIs

Private IPv4s per ENI

IPv6s per ENI

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmg6a.64xlarge

256

1024

64

24,000,000

32

31

10

1

600,000

32

Note

Instances of this instance family must use images in UEFI mode. If you use a custom image, ensure its boot mode is set to UEFI. For more information, see Instance boot mode.

General-purpose (enhanced balance) instance family ebmg6e

ebmg6e provides the following features:

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Third-party hypervisor compatibility for hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High packet rate scenarios, such as live video comments and telecom service forwarding.

    • Enterprise applications of all types and sizes.

    • Websites and application servers.

    • Game servers.

    • Small- and medium-sized database systems, caches, and search clusters.

    • Data analytics and compute.

    • Compute clusters and memory-intensive data processing.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of approximately 1:4.

    • Processor: Intel ® Xeon ® Platinum 8269CY (Cascade Lake) with a 2.5 GHz base frequency and a 3.2 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers a packet forwarding rate of up to 24,000,000 PPS.

The following table lists the instance types and specifications for the ebmg6e instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4s per ENI

IPv6s per ENI

Baseline disk IOPS

Baseline disk bandwidth (Gbit/s)

ecs.ebmg6e.26xlarge

104

384

32

24,000,000

1,800,000

32

10

1

480,000

16

General-purpose ECS Bare Metal Server family ebmg6

The ebmg6 instance family provides the following features:

  • Instance family overview: Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatibility with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Video encoding, decoding, and rendering.

    • Resource-intensive database applications for medium-to-large enterprises.

    • Compute clusters and memory-intensive data processing.

    • Data analytics and high-performance computing.

  • Compute:

    • vCPU-to-memory ratio of approximately 1:4.

    • Processor: Intel® Xeon® Platinum 8269CY (Cascade Lake) with a 2.5 GHz base frequency and a 3.2 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Provides high network performance with a network packet forwarding rate of 6,000,000 PPS.

The following table lists the instance types and specifications for the ebmg6 instance family.

Instance type

vCPU

Memory (GiB)

Baseline bandwidth (Gbit/s)

Packet rate (PPS)

Connections

ENIs

Private IPv4 addresses

IPv6 addresses

Baseline disk IOPS

Baseline disk bandwidth (Gbit/s)

ecs.ebmg6.26xlarge

104

384

32

6,000,000

1,800,000

32

20

1

200,000

16

Note

ECS Bare Metal Server instances do not provide basic CPU monitoring metrics by default. Install the CloudMonitor agent to obtain this data. For detailed instructions, see Install the CloudMonitor Agent.

Compute-optimized elastic bare metal family ebmc9ae

  • Instance family overview: Built on the CIPU architecture and equipped with the latest AMD EPYC Turin processors, this instance family provides stable compute performance, a powerful I/O engine, and chip-level security hardening.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Third-party hypervisor compatibility for hybrid cloud and multi-cloud deployment.

    • Containers, including but not limited to Docker, Clear Containers, and Pouch.

    • High network packet rate scenarios, such as live video comments and telecom service forwarding.

    • Web front-end servers.

    • Massively Multiplayer Online (MMO) game front-ends.

    • Data analytics, batch computing, and video encoding.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:2.

    • Processor: AMD EPYC Turin processors with a turbo frequency of up to 3.7 GHz. Instances in this family use physical cores to provide stable compute performance.

  • Storage:

  • Network:

The following table lists the instance types and specifications for the ebmc9ae instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (baseline/burst) (Gbit/s)

Network PPS

Connections

Queues

ENIs

Private IPv4 per ENI

IPv6 per ENI

Baseline cloud disk IOPS

Cloud disk bandwidth (baseline/burst) (Gbit/s)

ecs.ebmc9ae.48xlarge

192

384

100/None

30,000,000

6,000,000

64

38

50

50

600,000

50/None

Compute-optimized elastic bare metal server ebmc9i

  • Instance family overview: Powered by the new Alibaba Cloud CIPU architecture and Intel® Xeon® 6 processors with P-cores. This instance family delivers a more powerful I/O engine and chip-level security hardening.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatibility with third-party hypervisors for hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High packet rate scenarios, such as live video comments and telecom service forwarding.

    • Web front-end servers.

    • Front ends for massively multiplayer online (MMO) games.

    • Data analytics, batch processing, and video encoding.

    • High-performance scientific and engineering applications.

  • Compute:

    • Processor-to-memory ratio of 1:2.

    • Processor: Intel® Xeon® Granite Rapids with a base frequency of 3.2 GHz and an all-core turbo frequency of 3.6 GHz.

  • Storage:

  • Network:

The following table lists the instance types and metrics for the ebmc9i instance family.

Instance type

vCPUs

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Queues

ENIs

Private IPv4s/ENI

IPv6 addresses/ENI

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmc9i.48xlarge

192

384

64/None

20,000,000

6,000,000

64 (primary network interface)/16 (secondary network interface)

32

50

50

500,000/800,000

32/None

General-purpose ebmc8a instance family

  • Instance family overview: Powered by Alibaba Cloud's new CIPU architecture and AMD EPYC Genoa processors, this instance family delivers stable computing power, a more powerful I/O engine, and chip-level security hardening.

  • Use cases:

    • Workloads that need direct access to physical resources or require hardware-bound licenses.

    • Third-party hypervisor compatibility for hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High-packet-rate network applications, such as live video comments and telecommunication service forwarding.

    • Web front-end servers.

    • Front ends for massively multiplayer online (MMO) games.

    • Data analytics, batch processing, and video encoding.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:2.

    • Processor: AMD EPYC Genoa processors with a turbo frequency of up to 3.7 GHz, delivering stable computing power.

  • Storage:

  • Network:

The following table describes the instance types and specifications of the ebmc8a instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Queues

ENIs

Private IPv4 addresses

IPv6 addresses

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmc8a.48xlarge

192

384

64/None

18,000,000

3,000,000

64

38

30

30

500,000/None

32/None

Ebmc8y compute-optimized instance family

  • Instance family overview: Powered by the new Alibaba Cloud CIPU architecture, this instance family delivers stable computing power, a more powerful I/O engine, dedicated hardware resources, and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or have hardware-bound licensing requirements.

    • Compatible with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Video on demand and live streaming.

    • A wide range of enterprise applications.

    • Websites and application servers.

    • Data analytics and computing.

    • High-performance scientific and engineering applications.

  • Compute:

    • Processor: Alibaba Cloud's proprietary Yitian 710 processor, based on the ARM architecture, with a clock speed of at least 2.75 GHz. Hyper-threading is disabled for stable compute performance.

  • Storage:

  • Network:

Instance type

vCPU

Memory (GiB)

Network bandwidth (baseline/burst) (Gbit/s)

Network PPS (packets per second)

Connections

Multi-queue

Elastic network interfaces

Private IPv4 addresses per ENI

IPv6 addresses per ENI

Cloud disk IOPS (baseline/burst)

Cloud disk bandwidth (baseline/burst) (Gbit/s)

ecs.ebmc8y.32xlarge

128

256

64/None

20,000,000

3,000,000

64 (primary network interface)/32 (secondary network interface)

38

30

30

500,000/None

32/None

Ebmc8i instance family: Compute-optimized elastic bare metal server

  • Instance family overview: Powered by the new Alibaba Cloud CIPU architecture, this instance family provides stable computing power, an enhanced I/O engine, dedicated hardware resources, and physical isolation.

  • Use cases:

    • Workloads that need direct access to physical resources or require hardware-bound licenses.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployment.

    • Containers such as Docker, Clear Container, and Pouch.

    • High-packet-rate scenarios, such as live video commenting and telecommunications service forwarding.

    • Web front-end servers.

    • Front ends for massively multiplayer online (MMO) games.

    • Data analytics, batch processing, and video encoding.

    • High-performance scientific and engineering applications.

  • Compute:

    • Processor: Powered by Intel® Xeon® Emerald Rapids or Intel® Xeon® Sapphire Rapids processors with a base frequency of at least 2.7 GHz and an all-core turbo frequency of 3.2 GHz, providing stable computing power.

      Note

      When you purchase an instance of this family, the system randomly assigns one of the preceding processors. You cannot manually select a processor.

    • These instances support hyper-threading. Hyper-threading is enabled by default. For more information, see Change CPU options.

    • For information about operating system compatibility, see Intel instance types and operating system compatibility.

  • Storage:

  • Network:

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Multi-queue

ENIs

Private IPv4 addresses/ENI

IPv6 addresses/ENI

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmc8i.48xlarge

192

512

100/None

30,000,000

4,000,000

64 (primary network interface)/16 (secondary network interface)

72

30

30

1,000,000/None

48/None

Ebmc7 compute-optimized instance family

  • Instance family overview:

    • Built on the third-generation Dragon Architecture, this instance family uses chip-level fast path acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • It provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Compatible with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High network packet processing scenarios, such as live video comments and telecommunication service forwarding.

    • Web front-end servers.

    • Massively Multiplayer Online (MMO) game front-ends.

    • Data analytics, batch computing, and video encoding.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:2.

    • Processor: Intel® Xeon® Platinum 8369B (Ice Lake) with a 2.9 GHz base frequency and a 3.5 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For 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 network packet forwarding rate of 24,000,000 PPS.

The following table lists the specifications for the ebmc7 instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4 addresses

IPv6 addresses

Baseline disk IOPS

Baseline disk bandwidth (Gbit/s)

ecs.ebmc7.32xlarge

128

256

64

24,000,000

2,400,000

32

20

20

600,000

32

Compute-optimized ebmc7a instance family

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level acceleration to provide stable, predictable, and exceptionally high compute, storage, and network performance.

    • It provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Compatibility with third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Video encoding and decoding, and rendering.

    • Data analytics and compute.

  • Compute:

    • The vCPU-to-memory ratio is 1:2.

    • Processor: AMD EPYCTM MILAN processor with a 2.55 GHz base frequency and a single-core turbo frequency of up to 3.5 GHz, providing stable compute performance.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Offers high-performance networking with a packet processing rate of up to 24 million PPS.

The following table describes the instance types and specifications of the ebmc7a instance family.

Instance type

vCPU

Memory (GiB)

Network baseline bandwidth (Gbit/s)

Network PPS

Connections

Multi-queue

Elastic network interfaces

Private IPv4 addresses

IPv6 addresses

Baseline IOPS

Baseline bandwidth (Gbit/s)

ecs.ebmc7a.64xlarge

256

512

64

24 million

4 million

32

31

15

1

600,000

32

Note
  • ECS instances of this instance type require a UEFI boot mode. If you use a custom image, ensure it is configured for UEFI. For more information, see Instance boot mode.

  • The kernels of Ubuntu 18 and Debian 9 do not support the AMD EPYC MILAN processor. Do not create instances of this instance type from Ubuntu 18 or Debian 9 images, as instances created from them will fail to start.

Compute-optimized bare metal instance family ebmc6me

  • Instance family overview: Dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or use hardware-bound licenses.

    • Compatibility with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers (including but not limited to Docker, Clear Container, and Pouch).

    • Video encoding, decoding, and rendering.

    • Massively Multiplayer Online (MMO) game front-ends.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of approximately 1:3

    • Processor: Intel® Xeon® Gold 5218 (Cascade Lake) with a 2.3 GHz base frequency and up to 3.9 GHz turbo frequency

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Up to 6,000,000 network PPS.

The following table lists the specifications for the ebmc6me instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

IPv4s per ENI

IPv6s per ENI

Baseline IOPS

Baseline bandwidth (Gbit/s)

ecs.ebmc6me.16xlarge

64

192

32

6,000,000

1,800,000

32

10

1

200,000

16

Compute-optimized instance family ebmc6a

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family delivers stable, predictable, and ultra-high compute, storage, and network performance through chip-level fast path acceleration.

    • Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Hybrid cloud and multi-cloud deployments that use third-party hypervisors.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Video encoding and decoding, and rendering.

    • Data analytics and high-performance computing.

  • Compute:

    • vCPU-to-memory ratio of 1:2.

    • Processor: AMD EPYCTM ROME processors with a 2.6 GHz base frequency and a turbo frequency of up to 3.3 GHz.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Achieves a network PPS of 24 million.

The following table lists the specifications for the ebmc6a instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Queues

ENIs

Private IPv4 addresses

IPv6 addresses

Disk IOPS

Disk bandwidth (Gbit/s)

ecs.ebmc6a.64xlarge

256

512

64

24,000,000

32

31

10

1

600,000

32

Note

ECS instances of this instance family require an image that uses the UEFI boot mode. Custom images must use the UEFI boot mode. For more information, see Instance boot mode.

Compute-optimized ebmc6e instance family

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or use hardware-bound licenses.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High-rate network packet processing, such as live video comments and telecom service forwarding.

    • Web front-end servers.

    • Massively Multiplayer Online (MMO) game front-ends.

    • Data analytics, batch computing, and video encoding.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of approximately 1:2.

    • Processor: Intel ® Xeon ® Platinum 8269CY (Cascade Lake) with a 2.5 GHz base frequency and a 3.2 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Achieves a network PPS of 24 million.

The following table describes the instance types and specifications of the ebmc6e instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Elastic network interfaces

Private IPv4 addresses

IPv6 addresses

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmc6e.26xlarge

104

192

32

24,000,000

1,800,000

32

10

1

480,000

16

Compute-optimized instance family ebmc6

  • Instance family overview: Dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that need direct access to physical resources or require hardware-bound licenses.

    • Compatibility with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers (including but not limited to Docker, Clear Container, and Pouch).

    • Video encoding and decoding, and rendering.

    • Massively Multiplayer Online (MMO) game front-ends.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of approximately 1:2.

    • Processor: Intel® Xeon® Platinum 8269CY (Cascade Lake) with a 2.5 GHz base frequency and a 3.2 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Offers high network performance with a network packet forwarding rate of 6,000,000 PPS.

The following table lists the instance types and specifications of the ebmc6 instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4 addresses

IPv6 addresses

Baseline IOPS

Baseline bandwidth (Gbit/s)

ecs.ebmc6.26xlarge

104

192

32

6,000,000

1,800,000

32

20

1

200,000

16

Memory-optimized elastic bare metal server instance family ebmr9ae

  • Instance family overview: Powered by the new Alibaba Cloud CIPU architecture and the latest AMD EPYC Turin processors, this instance family provides stable computing power, a powerful I/O engine, and chip-level security.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Compatibility with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High-performance databases and in-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop, Spark clusters, and other enterprise applications that require large amounts of memory.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:8.

    • Processor: AMD EPYC Turin processors with a turbo frequency of up to 3.7 GHz. Instances in this family use physical cores to ensure stable compute performance.

  • Storage:

  • Network:

The following table lists the instance types and metrics for the ebmr9ae instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Multi-queue

ENIs

Private IPv4s per ENI

IPv6 addresses per ENI

Base disk IOPS

Disk bandwidth (Gbit/s)

ecs.ebmr9ae.48xlarge

192

1536

100/N/A

30,000,000

6,000,000

64

38

50

50

600,000

50/N/A

Memory-optimized elastic bare metal server ebmr9i

  • Instance family overview: Built on the new Alibaba Cloud CIPU architecture and powered by Intel® Xeon® 6 processors with P-cores, this instance family provides consistent performance, an enhanced I/O engine, and chip-level security hardening.

  • Use cases:

    • Workloads that need direct access to physical resources or require hardware-bound licenses.

    • Deployments that use third-party hypervisors in hybrid or multi-cloud environments.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High-performance databases and in-memory databases.

    • Data analytics and mining and distributed in-memory caches.

    • Hadoop, Spark clusters, and other enterprise applications that require large amounts of memory.

    • High-performance scientific and engineering applications.

  • Compute:

    • The vCPU-to-memory ratio is 1:8.

    • Processor: Intel® Xeon® Granite Rapids with a base frequency of 3.2 GHz and an all-core turbo frequency of 3.6 GHz.

  • Storage:

  • Network:

The following table lists the instance types and specifications for the ebmr9i instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Multi-queue

ENIs

IPv4 addresses per ENI

IPv6 addresses per ENI

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmr9i.48xlarge

192

1536

64/-

20,000,000

6,000,000

64 (primary)/16 (secondary)

32

50

50

500,000/800,000

32/-

General-purpose ebmr8a instance family

  • Instance type family overview: Powered by the new Alibaba Cloud CIPU architecture and AMD EPYC Genoa processors, this instance type family provides stable compute performance, a powerful I/O engine, and chip-level security hardening.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High-performance databases and in-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop and Spark clusters, and other enterprise applications that require large amounts of memory.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:8.

    • Processor: AMD EPYC Genoa processors with a turbo frequency of up to 3.7 GHz, which provides stable compute performance.

  • Storage:

  • Network:

The following table lists the specifications for the ebmr8a instance type family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s) (baseline/burst)

Network PPS

Connections

Queues

ENIs

IPv4s per ENI

IPv6s per ENI

Cloud disk IOPS (baseline/burst)

Cloud disk bandwidth (Gbit/s) (baseline/burst)

ecs.ebmr8a.48xlarge

192

1536

64/None

18,000,000

3,000,000

64

38

30

30

500,000/None

32/None

Memory-optimized Elastic Bare Metal Server instance family ebmr8y

Note

To use ebmr8y, submit a ticket to request access.

  • Instance family overview: Built on the new Alibaba Cloud CIPU architecture, this instance family delivers stable computing power, a powerful I/O engine, and dedicated hardware resources with physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Video on demand and live streaming.

    • Enterprise applications of all types and sizes.

    • Websites and application servers.

    • Data analytics and computing.

    • High-performance scientific and engineering applications.

  • Compute:

    • Processor: Alibaba Cloud's proprietary Yitian 710 ARM-based CPU with a clock speed of at least 2.75 GHz. Hyper-threading is disabled to ensure stable compute performance.

  • Storage:

  • Network:

The ebmr8y instance family offers the following specifications:

Instance type

vCPUs

Memory (GiB)

Network bandwidth (baseline/burst) (Gbit/s)

Network PPS

Connections

Multi-queue

Elastic network interfaces

Private IPv4 per ENI

IPv6 per ENI

Cloud disk IOPS (baseline/burst)

Cloud disk bandwidth (baseline/burst) (Gbit/s)

ecs.ebmr8y.32xlarge

128

1024

64/None

20,000,000

3,000,000

64 (primary network interface)/32 (secondary network interface)

38

30

30

500,000/None

32/None

Memory-optimized ebmr7 instance family

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family delivers stable, predictable, and ultra-high compute, storage, and network performance through chip-level acceleration.

    • These instances provide dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Supports third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High-performance databases and in-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop and Spark clusters, and other enterprise applications with high memory requirements.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:8.

    • Processor: Intel® Xeon® Platinum 8369B (Ice Lake) with a 2.9 GHz base frequency and a 3.5 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers ultra-high network performance, reaching 24,000,000 network PPS.

The following table lists specifications for the ebmr7 instance family.

Instance type

vCPUs

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4 addresses

IPv6 addresses

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmr7.32xlarge

128

1024

64

24,000,000

2,400,000

32

20

20

600,000

32

Memory-optimized ebmr7a instance family

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level acceleration to provide stable, predictable, and exceptional compute, storage, and network performance.

    • Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or have hardware-bound licenses.

    • Compatible with third-party hypervisors to support hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Container, and Pouch.

    • In-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop and Spark clusters, and other enterprise applications that require large amounts of memory.

  • Compute:

    • vCPU-to-memory ratio of 1:8

    • Processor: AMD EPYCTM MILAN processor with a 2.55 GHz base frequency and a single-core turbo frequency of up to 3.5 GHz, which ensures stable compute performance

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Exceptional network performance, reaching 24 million PPS.

The following table lists the specifications for the ebmr7a instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

Multi-queue

ENIs

Private IPv4s per ENI

IPv6s per ENI

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmr7a.64xlarge

256

2048

64

24,000,000

4,000,000

32

31

15

1

600,000

32

Note
  • Instances of this family must use images with the UEFI boot mode. If you use a custom image, ensure that its boot mode is set to UEFI. For more information, see Instance boot modes.

  • The kernels of Ubuntu 18 and Debian 9 do not support the AMD EPYCTM MILAN processor. Do not use images of these operating systems to create these instances, or they will fail to start.

Ebmr6a: Memory-optimized ECS Bare Metal Instance family

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • It provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Compatibility with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • In-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop and Spark clusters, and other enterprise applications that require large amounts of memory.

  • Compute:

    • vCPU-to-memory ratio of 1:8

    • Processor: AMD EPYCTM ROME processor with a 2.6 GHz base frequency and a turbo frequency of up to 3.3 GHz.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Supports a packet forwarding rate of up to 24,000,000 PPS.

The following table lists the specifications for the ebmr6a instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth

Network PPS

QP count

ENIs

ENI private IPv4s

ENI IPv6s

Baseline disk IOPS

Baseline disk bandwidth

ecs.ebmr6a.64xlarge

256

2048

64

24,000,000

32

31

10

1

600,000

32

Note

ECS instances of this instance family require an image that uses the UEFI boot mode. If you use a custom image, ensure its boot mode is set to UEFI. For more information, see Instance boot modes.

Memory-optimized elastic bare metal instance family ebmr6e

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatible with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers, including Docker, Clear Container, and Pouch.

    • High packet rate scenarios, such as live video comments and telecom service forwarding.

    • High-performance databases and in-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop and Spark clusters, and other enterprise applications that require large amounts of memory.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of approximately 1:8.

    • Processor: Intel® Xeon® Platinum 8269CY (Cascade Lake) with a 2.5 GHz base frequency and a 3.2 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers up to 24,000,000 network PPS.

The table below lists the instance types and specifications of the ebmr6e instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4 addresses per ENI

IPv6 addresses per ENI

Baseline cloud disk IOPS

Baseline cloud disk bandwidth (Gbit/s)

ecs.ebmr6e.26xlarge

104

768

32

24,000,000

1,800,000

32

10

1

480,000

16

Ebmr6: Memory-optimized elastic bare metal server family

  • Instance family overview: Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High-performance databases and in-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop and Spark clusters, and other enterprise applications that require large amounts of memory.

  • Compute:

    • vCPU-to-memory ratio of approximately 1:8.

    • Processor: Intel ® Xeon ® Platinum 8269CY (Cascade Lake) with a 2.5 GHz base frequency and a 3.2 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Packet forwarding rate (PPS) of up to 6,000,000.

The following table lists the instance types and specifications for the ebmr6 instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

Elastic network interfaces

Private IPv4 addresses per ENI

IPv6 addresses per ENI

Baseline cloud disk IOPS

Baseline cloud disk bandwidth (Gbit/s)

ecs.ebmr6.26xlarge

104

768

32

6,000,000

1,800,000

32

20

1

200,000

16

Persistent memory-optimized bare metal instance family (ebmre7p)

  • Instance family overview: Offers dedicated hardware resources and physical isolation.

  • Use cases:

    • In-memory databases, such as Redis.

    • High-performance databases, such as SAP HANA.

    • Other memory-intensive applications, such as AI and intelligent search applications.

  • Compute:

    • Powered by Intel® OptaneTM persistent memory, these instances are optimized for Redis applications, delivering excellent price-performance.

    • Supports up to 2,560 GiB of memory, including 512 GiB of DRAM and 2,048 GiB of Intel® OptaneTM persistent memory. The vCPU-to-memory ratio is approximately 1:20, making it ideal for memory-intensive applications.

    • Processor: Features 3rd Generation Intel® Xeon® Scalable (Ice Lake) processors with a base frequency of 2.7 GHz and an all-core turbo frequency of 3.5 GHz, delivering stable compute performance.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Features high network performance, with a packet forwarding rate (PPS) of 24,000,000.

The following table lists the instance types and specifications of the ebmre7p instance family.

Instance type

vCPU

Memory (GiB)

Persistent memory (GiB)

Baseline network bandwidth (Gbit/s)

Packet forwarding rate (PPS)

ENIs

Private IPv4 per ENI

IPv6 per ENI

Baseline disk IOPS

Baseline disk bandwidth (Gbit/s)

ecs.ebmre7p.32xlarge

128

512

2048

64

24,000,000

32

15

1

600,000

32

Persistent memory-optimized bare metal instance family ebmre6p

To access ebmre6p, submit a ticket.

  • Instance family overview: Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • In-memory databases, such as Redis.

    • High-performance databases, such as SAP HANA.

    • Other memory-intensive applications, such as AI and intelligent search applications.

  • Compute:

    • Powered by Intel® OptaneTM persistent memory, these instances are fully optimized for Redis applications to deliver excellent price-performance.

    • Supports up to 1,920 GiB of memory (384 GiB of DRAM and 1,536 GiB of Intel® OptaneTM persistent memory). The vCPU-to-memory ratio is approximately 1:20, making these instances ideal for memory-intensive applications.

    • Processor: 2.5 GHz Intel® Xeon® Platinum 8269CY (Cascade Lake) with an all-core turbo frequency of 3.2 GHz, delivering stable compute performance.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • High network performance, with a packet forwarding rate of up to 6,000,000 PPS.

The following table lists the instance types and performance metrics for the ebmre6p instance family.

Instance type

vCPU

Memory (GiB)

Persistent memory (GiB)

Network bandwidth (Gbit/s)

Packet forwarding rate (PPS)

ENIs

Private IPv4 addresses

IPv6 addresses

Cloud disk IOPS

Cloud disk bandwidth (Gbit/s)

ecs.ebmre6p.26xlarge

104

384

1536

32

6,000,000

31

10

1

200,000

16

Memory-optimized instance family ebmre6-6t

To use the ebmre6-6t instance family, submit a ticket.

  • Instance family overview: Offers dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • High-performance databases and in-memory databases (such as SAP HANA).

    • Memory-intensive applications.

    • Big data processing engines (such as Apache Spark and Presto).

  • Compute:

    • vCPU-to-memory ratio of approximately 1:30.

    • Processor: Intel® Xeon® Platinum 8269 (Cascade Lake) with a 2.5 GHz base frequency and a 3.2 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers a network packet processing capability of 6,000,000 PPS.

The following table details the instance types and metrics for the ebmre6-6t instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4 addresses per ENI

IPv6 addresses per ENI

Baseline cloud disk IOPS

Baseline cloud disk bandwidth (Gbit/s)

ecs.ebmre6-6t.52xlarge

208

6144

32

6,000,000

1,800,000

32

10

1

200,000

16

High-frequency general-purpose EBM family ebmhfg7

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • High packet rate scenarios, such as live video comments and telecom service forwarding.

    • Enterprise applications of all types and sizes.

    • Game servers.

    • Small and medium-sized database systems, caches, and search clusters.

    • High-performance scientific computing.

    • Video encoding applications.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: 3rd Generation Intel® Xeon® Scalable processors (Cooper Lake architecture), with a minimum base frequency of 3.3 GHz and an all-core turbo frequency of 3.8 GHz.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Achieves ultra-high network performance of 24,000,000 network PPS.

The following table lists the instance types and specifications for the ebmhfg7 instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Multi-queue

ENIs

Private IPv4 addresses

IPv6 addresses

Cloud disk IOPS

Disk bandwidth (Gbit/s)

ecs.ebmhfg7.48xlarge

192

768

64

24,000,000

32

31

10

1

600,000

32

High-frequency compute-optimized instance family ebmhfc7

  • Instance family overview:

    • Built on the third-generation X-Dragon architecture, this instance family uses chip-level acceleration to deliver stable, predictable, and high compute, storage, and network performance.

    • It provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • High packet rate scenarios, such as live video comments and telecommunication service forwarding.

    • High-performance front-end server clusters.

    • Massively Multiplayer Online (MMO) game front-ends.

    • Data analytics, batch computing, and video encoding.

    • High-performance scientific and engineering applications.

  • Compute:

    • vCPU-to-memory ratio of 1:2.

    • Processor: 3rd Generation Intel® Xeon® Scalable processors (Cooper Lake) with a base frequency of at least 3.3 GHz and an all-core turbo frequency of 3.8 GHz.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Provides high network performance of up to 24,000,000 packets per second (PPS).

The following table lists the instance types and specifications for the ebmhfc7 instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Queues

ENIs

Private IPv4 addresses

IPv6 addresses

Baseline IOPS

Baseline disk bandwidth (Gbit/s)

ecs.ebmhfc7.48xlarge

192

384

64

24,000,000

32

31

10

1

600,000

32

ebmhfr7: High-frequency memory-optimized bare metal instance family

  • Instance family introduction:

    • Powered by the 3rd-generation X-Dragon architecture, these instances use chip-level fast path acceleration to deliver stable, predictable, and ultra-high compute, storage, and network performance.

    • They offer dedicated hardware resources and physical isolation.

  • Scenarios:

    • Workloads that require direct access to physical resources or have hardware-bound licenses.

    • High packet rate scenarios, such as live video comments and telecom service forwarding.

    • High-performance databases and in-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop and Spark clusters, and other enterprise applications with high memory requirements.

  • Compute:

    • vCPU-to-memory ratio of 1:8.

    • Processor: 3rd Gen Intel® Xeon® Scalable processors (Cooper Lake) with a base frequency of at least 3.3 GHz and an all-core turbo frequency of 3.8 GHz.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers ultra-high network performance of up to 24 million PPS.

The following table lists the specifications of instance types in the ebmhfr7 family.

Instance type

vCPUs

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Queues

ENIs

Private IPv4 addresses

IPv6 addresses

Baseline IOPS

Baseline bandwidth (Gbit/s)

ecs.ebmhfr7.48xlarge

192

1536

64

24,000,000

32

31

10

1

600,000

32

Ebmhfg6: High-frequency general-purpose bare metal instance family

  • Instance family overview: Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or are tied to specific hardware for licensing.

    • Compatible with third-party hypervisors to support hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Heavy-duty database applications for medium-to-large enterprises.

    • Video encoding, decoding, and rendering.

  • Compute:

    • vCPU-to-memory ratio of 1:4.8.

    • Processor: Intel® Xeon® Platinum 8269CY (Cascade Lake) with a 3.1 GHz base frequency and a 3.5 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Features a high Network PPS of 6,000,000.

The following table details the specifications for each instance type in the ebmhfg6 instance family.

Instance type

vCPU

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4 addresses

IPv6 addresses

Disk IOPS

Disk bandwidth (Gbit/s)

ecs.ebmhfg6.20xlarge

80

384

32

6,000,000

1,800,000

32

20

1

200,000

16

High-frequency compute-optimized elastic bare metal server family ebmhfc6

  • Instance family overview: Offers dedicated hardware resources and physical isolation.

  • Use cases:

    • High packet rate scenarios, such as live video comments and telecom service forwarding.

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployment.

    • Containers, such as Docker, Clear Container, and Pouch.

    • Video encoding and decoding, and rendering.

  • Compute:

    • vCPU-to-memory ratio of 1:2.4

    • Processor: Intel ® Xeon ® Platinum 8269CY (Cascade Lake) with a 3.1 GHz base frequency and a 3.5 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • High network performance with a packet forwarding rate of up to 6,000,000 PPS.

The following table lists the specifications for the ebmhfc6 instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4s per ENI

IPv6s per ENI

Baseline cloud disk IOPS

Baseline cloud disk bandwidth (Gbit/s)

ecs.ebmhfc6.20xlarge

80

192

32

6,000,000

1,800,000

32

20

1

200,000

16

High-frequency memory-optimized instance family ebmhfr6

  • Instance family overview: Offers dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers (such as Docker, Clear Container, and Pouch).

    • High-performance databases and in-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop and Spark clusters, and other enterprise applications that require large memory.

  • Compute:

    • vCPU-to-memory ratio of 1:9.6.

    • Processor: Intel® Xeon® Platinum 8269CY (Cascade Lake) with a 3.1 GHz base frequency and a 3.5 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Offers a packet forwarding rate (PPS) of 6,000,000.

The following table lists the instance types and specifications for the ebmhfr6 instance family.

Instance type

vCPU

Memory (GiB)

Base network bandwidth (Gbit/s)

Packet forwarding rate (PPS)

Connections

ENIs

ENI private IPv4s

ENI IPv6s

Baseline IOPS

Baseline bandwidth (Gbit/s)

ecs.ebmhfr6.20xlarge

80

768

32

6,000,000

1,800,000

32

20

1

200,000

16

GPU-accelerated elastic bare metal server instance family ebmgn9g

Important

ebmgn9g is in invitation-only preview. To request access, submit a ticket.

  • Instance family overview: ebmgn9g is the 9th generation of Alibaba Cloud's cost-effective GPU bare metal instance family. Powered by the latest CIPU 2.0, these instances feature high-frequency CPUs, large-capacity memory, and the new Blackwell architecture professional GPUs. They deliver cost-effective GPU cloud services for various GPU-accelerated scenarios, including autonomous driving and embodied intelligence training, large model inference, film and animation rendering, and metaverse and cloud gaming services.

  • Use cases and product highlights:

    • Autonomous driving and embodied intelligence:
      These instances provide 256 vCPUs running at an all-core frequency of over 4.2 GHz, paired with 2.3 TB of memory to support the data processing demands of autonomous driving and embodied intelligence training workloads.

    • Search and recommendation:
      The Blackwell GPU delivers 123 TFLOPS of TF32 compute. Each GPU is paired with an average of 32 vCPUs and 153 GB/s of memory bandwidth, providing an optimal configuration for search and advertising workloads.

    • Large model inference:
      This new-generation GPU delivers a significant leap in compute power over its predecessor. It features an increased GPU memory bandwidth of 1,344 GB/s and new support for FP4, which enhances both inference performance and cost-effectiveness. The eight GPUs are interconnected via PCIe Gen5, providing a bandwidth of 128 GB/s and dramatically improving multi-card parallel inference efficiency.

    • Cloud gaming, rendering, and metaverse:
      With processor frequencies reaching up to 5 GHz, these instances are ideal for 3D modeling. The GPUs offer native graphics capabilities and are supported by workstation-grade graphics drivers certified for professional design. With full OpenGL acceleration, they are an excellent choice for high-end film and animation development and CAD design.

  • Powered by the latest CIPU 2.0 cloud processor:

    The 2nd-generation CIPU provides higher cloud processing performance and delivers stronger compute offload for components such as eRDMA, VPC, and EBS. Bare metal instances provide direct access to physical resources and are ideal for workloads that require hardware-bound licenses. They support various containers, such as Docker, Clear Container, and Pouch.

  • Compute:

    • Powered by new Blackwell architecture professional GPUs:

      • They support professional-grade graphics processing with OpenGL.

      • They support popular acceleration features like RTX and TensorRT and now support FP4 and PCIe Gen5 interconnect.

      • They use a PCIe switch interconnect, which improves NCCL performance by 36% compared to a direct-to-CPU connection. This can boost performance by up to 9% for multi-card sharded large model inference.

    • Key GPU parameters:

      GPU architecture

      GPU memory

      Compute performance

      Video encoding/decoding

      Inter-GPU interconnect

      Acceleration APIs

      Blackwell

      • Capacity: 48 GB

      • Bandwidth: 1,344 GB/s

      • TF32: 123 TFLOPS

      • FP32: 52 TFLOPS

      • FP16/BF16: 261 TFLOPS

      • FP8/INT8: 533 TFLOPS

      • FP4: 970 TFLOPS

      • RT core: 196 TFLOPS

      • 3 × Video Encoder

      • 3 × Video Decoder

      • PCIe Gen5 x16: 128 GB/s

      • Supports P2P

      Supports DX12,

      OpenGL 4.6, Vulkan 1.3, CUDA 12.8, OpenCL 3.0, and DirectCompute

    • Processor: Instances are powered by AMD Turin-C processors with a frequency of 3.3 GHz to 5 GHz and an all-core turbo frequency of up to 4.2 GHz.

  • Storage:

  • Network:

    • Supports IPv4 and IPv6. For more information about IPv6 communication, see IPv6 Communication.

    • These instances deliver ultra-high network performance with a packet forwarding rate of up to 30,000,000 PPS.

    • These instances support Elastic RDMA Interface (ERI), which enables RDMA passthrough acceleration over VPC networks. This boosts bandwidth to 360 Gbit/s and is ideal for training workloads in autonomous driving, embodied intelligence, computer vision (CV), and traditional models.

The following table lists the specifications for the ebmgn9g instance family.

Instance specification

vCPU

Memory (GiB)

GPU memory

Network bandwidth (Gbit/s)

Network PPS

Private IPv4 addresses per ENI

IPv6 addresses per ENI

Multi-queue (primary/secondary)

ENIs

Max data disks

Max cloud disk bandwidth (GB/s)

ecs.ebmgn9g.64xlarge

256

2304

48 GB × 8

360 (180 × 2)

30,000,000

30

30

64/16

38

33

8

Note

Instances in the ebmgn9g instance family require an image with the UEFI boot mode. If you use a custom image, ensure that it supports the UEFI boot mode and that its boot mode is set to UEFI. For more information, see Instance boot modes.

GPU-accelerated elastic bare metal server family ebmgn9ge

Important

The ebmgn9ge instance family is in invitation-only preview. To request access, please submit a ticket.

  • Instance family overview: The ebmgn9ge instance family is the 9th generation of Alibaba Cloud's cost-effective GPU bare metal instances. Powered by the latest CIPU 2.0, these instances combine high-frequency CPUs, large-capacity memory, and professional GPUs based on the new Blackwell architecture. They deliver cost-effective GPU cloud services for a wide range of GPU-accelerated scenarios, including autonomous driving and embodied intelligence training, large model inference, film and animation rendering, and metaverse or cloud gaming services.

  • Use cases and key features:

    • Autonomous driving and embodied intelligence:
      These instances provide 256 vCPUs with an all-core turbo frequency of over 4.2 GHz, paired with 2.3 TB of memory to support the data processing demands of autonomous driving and embodied intelligence training workloads.

    • Search and recommendation:
      The Blackwell GPU delivers 126 TFLOPS of TF32 computing power. Each GPU is paired with an average of 32 vCPUs and 153 GB/s of memory bandwidth, providing an optimal configuration for search and advertising workloads.

    • Large model inference:

      The ebmgn9ge instance family is specially designed for large language models (LLMs), offering 72 GB of GPU memory per card and a GPU memory bandwidth of 1,344 GB/s. This delivers high-performance inference for LLMs. With the new FP4 compute architecture and 128 GB/s of PCIe Gen5 bandwidth, these instances can support 8-GPU parallel inference for models with over 671B parameters.

    • Cloud gaming, rendering, and metaverse:
      With CPU frequencies up to 5 GHz, these instances excel at 3D modeling. The GPUs provide native graphics capabilities and are supported by workstation-grade graphics drivers certified for professional design. Full OpenGL acceleration makes them ideal for high-end film and animation development and CAD design.

  • Powered by the latest CIPU 2.0:

    The second-generation CIPU improves processing power and computing capabilities for eRDMA, VPC, and EBS components. These bare metal instances are ideal for workloads that require direct access to physical resources or hardware-bound licenses. They also support various containers, such as Docker, Clear Container, and Pouch.

  • Compute:

    • Powered by new Blackwell architecture professional GPUs:

      • Supports professional-grade graphics processing with OpenGL.

      • In addition to popular acceleration features like RTX and TensorRT, the GPU now supports FP4 and PCIe Gen5 interconnect.

      • A PCIe switch interconnect improves NCCL performance by 36% compared to a direct-to-CPU connection. This can boost performance by up to 9% for multi-GPU sharded large model inference.

    • Key GPU specifications:

      GPU architecture

      GPU memory

      Compute performance

      Video encoding/decoding

      GPU interconnect

      Acceleration APIs

      Blackwell

      • Capacity: 72 GB

      • Bandwidth: 1,344 GB/s

      • TF32: 126 TFLOPS

      • FP32: 52 TFLOPS

      • FP16/BF16: 266 TFLOPS

      • FP8/INT8: 530 TFLOPS

      • FP4: 971 TFLOPS

      • RT Core: 196 TFLOPS

      • 3 × Video Encoder

      • 3 × Video Decoder

      • PCIe Gen5 x16: 128 GB/s

      • P2P supported

      Supports DX12,

      OpenGL 4.6, Vulkan 1.3, CUDA 12.8, OpenCL 3.0, and DirectCompute

    • Processor: AMD Turin-C processors with a frequency range of 3.3 GHz to 5 GHz and an all-core turbo frequency of up to 4.2 GHz.

  • Storage:

  • Network:

    • Supports IPv4 and IPv6. For more information about IPv6 communication, see IPv6 Communication.

    • The instance supports a packet forwarding rate of up to 30,000,000 PPS.

    • Supports Elastic RDMA Interface (ERI), which enables RDMA passthrough acceleration over VPC networks. This boosts bandwidth to 360 Gbit/s and is well-suited for training workloads in autonomous driving, embodied intelligence, computer vision (CV), and traditional models.

      Note

      For instructions on how to use ERI, see Enable eRDMA on enterprise-level instances or Enable eRDMA on GPU instances.

The following table lists the instance types and key specifications for the ebmgn9ge instance family.

Note

For a lower-cost version with less memory, consider ebmgn9gc.

Instance type

vCPUs

Memory (GiB)

GPU memory

Baseline network bandwidth (Gbit/s)

Packet rate (PPS)

Private IPv4s per ENI

IPv6s per ENI

Multi-queue

ENIs

Max data disks

Max disk bandwidth (GB/s)

ecs.ebmgn9ge.64xlarge

256

2,304

72 GB × 8

360 (180 × 2)

30,000,000

30

30

64/16

38

33

8

Note

The ebmgn9ge instance type requires images that use the UEFI boot mode. If you use a custom image, ensure that the image supports UEFI boot mode and its boot mode property is set to UEFI mode. For more information, see Instance boot mode.

GPU-accelerated elastic bare metal server instance family ebmgn9gc

Important

The ebmgn9gc instance family is in beta testing. To request access, submit a ticket.

  • Instance family overview: ebmgn9gc is the 9th generation of Alibaba Cloud's GPU bare metal instances. Powered by the latest CIPU 2.0, these instances combine high-frequency CPUs, large memory capacity, and the new Blackwell architecture professional GPUs. These instances are ideal for a wide range of GPU-accelerated workloads, such as autonomous driving and embodied intelligence training, large model inference, film and animation rendering, and metaverse or cloud gaming services.

  • Use cases and product highlights:

    • Autonomous driving and embodied intelligence:
      Features 256 vCPUs with a CPU frequency of 3.3 GHz to 5.0 GHz and a stable all-core frequency of over 4.2 GHz. Paired with 1.5 TB of memory, it handles the data processing requirements for autonomous driving and embodied intelligence training workloads.

    • Search and recommendation:
      The Blackwell GPU delivers 126 TFLOPS of high-performance TF32 computing power. Each GPU is paired with an average of 32 vCPUs and 153 GB/s of memory bandwidth, creating an optimal configuration for search and advertising workloads.

    • Large model inference:

      The ebmgn9gc instance family is specially designed for large language models (LLMs). Each card offers 72 GB of GPU memory and a GPU memory bandwidth of 1,344 GB/s to provide high-performance computing power for LLM inference. Combined with the new FP4 compute architecture and 128 GB/s of PCIe Gen5 bandwidth, it can support 8-card parallel inference for models with more than 671 billion parameters.

    • Cloud gaming, rendering, and metaverse:
      With CPU frequencies up to 5.0 GHz, these instances are ideal for 3D modeling. The GPUs provide native graphics capabilities and are supported by workstation-grade graphics drivers certified for professional design. With full acceleration for OpenGL, they are the optimal choice for high-end film and animation development and CAD design.

  • Powered by the latest CIPU 2.0:

    The 2nd generation CIPU delivers enhanced processing and compute capabilities for eRDMA, VPC, and EBS components. These bare metal instances are ideal for workloads that require direct access to physical resources or hardware-bound licenses. They also support various containers, such as Docker, Clear Container, and Pouch.

  • Compute:

    • Powered by new Blackwell architecture professional GPUs:

      • Supports professional-grade graphics processing with OpenGL.

      • Supports popular acceleration features such as RTX and TensorRT, and now supports FP4 and PCIe Gen5 interconnect.

      • A PCIe switch interconnect improves NCCL performance by 36% compared to a direct-to-CPU connection. This can boost performance by up to 9% for multi-card sharded large model inference.

    • GPU key parameters:

      GPU architecture

      GPU memory

      Compute performance

      Video encoding/decoding

      GPU-to-GPU interconnect

      Acceleration APIs

      Blackwell

      • Capacity: 72 GB

      • Bandwidth: 1,344 GB/s

      • TF32: 126 TFLOPS

      • FP32: 52 TFLOPS

      • FP16/BF16: 266 TFLOPS

      • FP8/INT8: 530 TFLOPS

      • FP4: 971 TFLOPS

      • RT Core: 196 TFLOPS

      • 3 × Video Encoders

      • 3 × Video Decoders

      • PCIe Gen5 x16: 128 GB/s

      • P2P supported

      Supports DX12,

      OpenGL 4.6, Vulkan 1.3, CUDA 12.8, OpenCL 3.0, and DirectCompute

    • Processor: AMD Turin-C processors with a frequency of 3.3 GHz to 5.0 GHz and an all-core turbo frequency up to 4.2 GHz.

  • Storage:

  • Network:

    • Supports IPv4 and IPv6. For more information about IPv6 communication, see IPv6 communication.

    • Achieves a packet forwarding rate of up to 30,000,000 PPS.

    • Supports Elastic RDMA Interface (ERI), which enables RDMA passthrough acceleration over VPC networks. This boosts bandwidth to 360 Gbit/s and is ideal for training workloads in autonomous driving, embodied intelligence, computer vision (CV), and traditional models.

The following table lists the instance types of the ebmgn9gc family.

Note

For an option with more memory, consider ebmgn9ge.

Instance type

vCPUs

Memory (GiB)

GPU memory

Network bandwidth (Gbit/s)

Packet rate (PPS)

Private IPv4s per ENI

IPv6s per ENI

Queue pairs

Maximum ENIs

Maximum data disks

Cloud disk bandwidth (GB/s)

ecs.ebmgn9gc.64xlarge

256

1,536

72 GB × 8

360 (180 × 2)

30,000,000

30

30

64/16

38

33

8

Note

Instances of the ebmgn9gc family require an image with UEFI boot mode. If you use a custom image, ensure it supports UEFI boot mode and its boot mode property is set to UEFI. For more information, see Instance boot mode.

ebmgn8v: GPU elastic bare metal instance family

This instance family is currently available only in select regions. To request access, please contact your Alibaba Cloud sales representative.

  • Instance family overview: The ebmgn8v is Alibaba Cloud's 8th-generation accelerated computing instance family (an Elastic Bare Metal Server instance family) designed for AI model training and large-parameter models. Each instance is a bare metal server equipped with eight GPUs.

  • Use cases:

    • Cost-effective for multi-GPU parallel inference on large language models (LLMs) with over 70 billion parameters.

    • A single GPU provides 39.5 TFLOPS of FP32 computing power and has outstanding performance in traditional AI model training and autonomous driving training workloads.

    • The eight GPUs are interconnected via NVLink, making these instances suitable for training small- to medium-sized models.

  • Features and positioning:

    • High-speed, large-capacity GPU memory: Each GPU provides 96 GB of HBM3 memory with a bandwidth of 4 TB/s, accelerating model training and inference.

    • High inter-GPU bandwidth: 900 GB/s NVLink interconnects the GPUs, delivering multi-GPU training and inference efficiency that far exceeds previous generations.

    • Large model quantization technology: FP8 support optimizes training and inference for large-parameter models, improving computational speed while reducing GPU memory usage.

  • Compute:

    • Powered by the latest Cloud Processor Unit (CIPU) 1.0:

      • It decouples compute and storage, allowing you to flexibly select the required storage resources. This design increases inter-instance bandwidth to 160 Gbit/s compared to 7th-generation GPU instances, enabling faster data transfer and processing.

      • Unlike traditional virtualized instances, the CIPU provides bare metal capabilities that support P2P communication between GPU instances.

    • These instances are powered by 4th Generation Intel® Xeon® Scalable processors and provide 192 vCPUs with an all-core turbo frequency of up to 3.1 GHz.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Ultra-high network performance with a packet forwarding rate of 30 million PPS.

    • These instances support Elastic RDMA Interface (ERI), which enables RDMA passthrough acceleration over a VPC network, boosting bandwidth to 160 Gbit/s for training computer vision (CV) and traditional models.

      Note

      For instructions on how to use ERI, see Enable eRDMA on enterprise-level instances or Enable eRDMA on GPU instances.

The following table lists the instance types and their specifications for the ebmgn8v instance family.

Instance type

vCPUs

Memory (GiB)

GPU memory

Baseline network bandwidth (Gbit/s)

Packet forwarding rate (PPS)

Private IPv4 addresses per ENI

IPv6 addresses per ENI

Queue-pairs per ENI (primary and secondary)

Elastic network interfaces (ENIs)

Maximum data disks

Maximum cloud disk bandwidth (GB/s)

ecs.ebmgn8v.48xlarge

192

1,024

96 GB × 8

170 (85 × 2)

30,000,000

30

30

64

32

31

6

Note

ebmgn8v instances must be launched from an image that uses the UEFI boot mode. If you use a custom image, ensure it supports UEFI boot mode and its boot mode property is set to UEFI. For more information, see Instance launch mode.

GPU-accelerated elastic bare metal instance family ebmgn8ia

This instance is currently available only in select international regions. To request access, please contact your Alibaba Cloud sales representative.

  • Instance family overview: The ebmgn8ia instance family is Alibaba Cloud's 8th-generation, GPU-accelerated elastic bare metal server instance family. It is designed for sparse GPU computing workloads that require a high vCPU-to-GPU ratio, such as search, recommendation, and simulation. Powered by the latest NVIDIA L20 GPUs, each instance is a bare metal host with two high-frequency CPUs and four GPUs.

  • Key features and use cases:

    • High frequency: Powered by two AMD EPYC™ Genoa 9T34 processors, each with 64 physical cores, these instances deliver a total of 256 vCPUs and operate at a high frequency of 3.4 GHz to 3.75 GHz. This boost in single-core performance is ideal for CAD modeling and speeds up pre-processing in CAE simulations.

    • Sparse resource ratio: On average, each GPU has 64 vCPUs and 384 GiB of memory, providing a memory bandwidth of 230 GB/s. This configuration is ideal for GPU computing scenarios that require high I/O throughput, such as advertising, search, recommendation, traditional CAE simulations, and some CPU-based rendering in film production.

  • Powered by the latest CIPU 1.0 cloud processor:

    • It decouples compute and storage, which lets you flexibly select storage resources. Compared to the previous generation, the inter-node bandwidth for this instance type increases to 160 Gbit/s, which allows for faster data transfer and processing.

    • The CIPU provides bare metal capabilities that support PCIe P2P communication between GPU instances, an advantage over traditional virtualized instances.

  • Compute:

    • Features new NVIDIA L20 enterprise-grade GPUs:

      • Supports common acceleration features such as vGPU, RTX, and TensorRT.

      • Supports FP8 precision to improve computing efficiency.

    • Key NVIDIA L20 specifications:

      GPU architecture

      GPU memory

      Compute performance

      Video encoding/decoding

      GPU-to-GPU interconnect

      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 Encoder (+AV1)

      • 3 × Video Decoder

      • 4 × JPEG Decoder

      • PCIe: PCIe Gen4 x16

      • Bandwidth: 64 GB/s

    • Processor: AMD EPYC™ Genoa 9T34 processors with a frequency of 3.4 GHz to 3.75 GHz.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers ultra-high network performance with a packet forwarding rate of up to 30,000,000 PPS.

    • Supports eRDMA (Elastic RDMA Interface), which enables RDMA passthrough acceleration over a VPC network. This boosts bandwidth to 160 Gbit/s and is ideal for training computer vision (CV) and traditional models.

      Note

      For instructions on how to use eRDMA, see Enable eRDMA on enterprise-level instances or Enable eRDMA on GPU instances.

The following table lists the instance specifications for the ebmgn8ia instance family.

Instance type

vCPUs

Memory (GiB)

GPU

GPU memory

Baseline bandwidth (Gbit/s)

Network PPS

Private IPv4 per ENI

IPv6 per ENI

QPs (primary/secondary)

ENIs

Max data disks

Max disk bandwidth (GB/s)

ecs.ebmgn8ia.64xlarge

256

1536

L20 × 4

48 GB × 4

160 (80 × 2)

30,000,000

30

30

64/16

32

31

6

Note

ebmgn8ia instances must be launched from an image that uses the UEFI boot mode. If you use a custom image, ensure that it supports UEFI boot mode and its boot mode property is set to UEFI. For more information, see Instance boot modes.

GPU-accelerated elastic bare metal instance family: ebmgn8is

This instance is only available in select international regions. To request access, please contact your Alibaba Cloud sales representative.

  • Instance family overview: The ebmgn8is, an elastic bare metal server instance family, is Alibaba Cloud's 8th-generation accelerated computing family, designed for the growing demand of generative AI.

  • Product highlights and positioning:

    • Graphics processing: Powered by high-frequency 4th Generation Intel® Xeon® Scalable processors, these instances provide the CPU power for demanding 3D modeling, ensuring a smooth design and rendering experience.

    • Inference tasks: Accelerate inference tasks with the new NVIDIA L20 GPUs, each featuring 48 GB of GPU memory and support for the FP8 floating-point format. When paired with ACK, these instances flexibly support inference on various AIGC models and are especially well-suited for LLM inference tasks with fewer than 70 billion parameters.

    • Training tasks: These instances deliver cost-effective computing power and offer double the FP32 performance of 7th-generation inference instances. This makes them ideal for training small to medium-sized models, such as computer vision (CV) models that use FP32.

  • Use cases:

    • Use GRID drivers with images from Alibaba Cloud Marketplace to enable OpenGL and Direct3D graphics, providing workstation-grade graphics processing for animation, film and television VFX production, and rendering.

    • Use ACK to cost-effectively support AIGC image generation and large language model (LLM) inference (for models with fewer than 70 billion parameters).

    • Other general AI recognition workloads, such as image and speech recognition.

  • Powered by the latest CIPU 1.0 (Cloud-infrastructure Processing Unit):

    • Decouples compute and storage, allowing you to flexibly select the storage resources you need. Compared to the previous generation, the inter-node bandwidth for this instance family is increased to 160 Gbit/s for faster data transfer and processing.

    • Unlike traditional virtualized instances, the CIPU enables direct PCIe P2P communication between GPUs.

  • Compute:

    • Powered by new NVIDIA L20 enterprise-grade GPUs:

      • Supports common acceleration features such as vGPU, RTX, and TensorRT.

      • It uses a PCIe switch interconnect, improving NCCL performance by 36% compared to a direct-to-CPU connection. This can boost inference performance by up to 9% for multi-GPU sharded large model inference.

    • NVIDIA L20 key parameters:

      GPU architecture

      GPU memory

      Compute performance

      Video encoders/decoders

      GPU-to-GPU interconnect

      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 Encoder (+AV1)

      • 3 × Video Decoder

      • 4 × JPEG Decoder

      • PCIe Interface: PCIe Gen4 x16

      • Bandwidth: 64 GB/s

    • Processor: Intel® Xeon® Scalable processors (SPR) with a 3.4 GHz base frequency and an all-core turbo frequency of up to 3.9 GHz.

  • Storage:

  • Network:

The following table lists the instance specifications for the ebmgn8is instance family.

Instance type

vCPU

Memory (GiB)

GPU

GPU memory

Baseline network bandwidth (Gbit/s)

Network PPS

Private IPv4 addresses

IPv6 addresses

Queues (primary/secondary ENI)

ENIs

Max data disks

Max cloud disk bandwidth (GB/s)

ecs.ebmgn8is.32xlarge

128

1024

L20 × 8

48 GB × 8

160 (80 × 2)

30,000,000

30

30

64/16

32

31

6

Note

Instances in the ebmgn8is family require images that use the UEFI boot mode. If you use a custom image, ensure that it supports UEFI boot mode and that its boot mode property is set to UEFI. For more information, see Instance boot mode.

GPU compute-optimized bare metal instance family ebmgn7ex

  • Instance family overview: The ebmgn7ex is a high-bandwidth instance family from Alibaba Cloud, designed for the growing demands of large-scale AI training. Each instance is a bare metal server. Built on the fourth-generation SHENLONG architecture and Alibaba Cloud's new CIPU architecture, ebmgn7ex instances can interconnect over an eRDMA network, achieving RDMA communication at a bandwidth of 160 Gbit/s. By enabling eRDMA, you can elastically scale the instances in your cluster to meet your large-scale AI training needs.

  • Use cases:

    • Deep learning training and development.

    • High-performance computing (HPC) acceleration and simulation.

    Important

    When you use AI training services that feature a high communication load, such as transformer models, you must enable NVLink for GPU-to-GPU communication. Otherwise, data may be damaged due to unpredictable failures that are caused by large-scale data transmission over Peripheral Component Interconnect Express (PCIe) links. If you do not understand the topology of the communication links that are used for AI training services, submit a ticket to obtain technical support.

  • Compute:

    • Processor: 3rd Generation Intel® Xeon® Scalable processors (Icelake) with a base frequency of 2.9 GHz and an all-core turbo frequency of 3.5 GHz. The processors support the PCIe 4.0 interface.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Supports physical NICs.

    • Achieves a packet forwarding rate of up to 24,000,000 pps.

    • Supports the Elastic RDMA Interface (ERI) for RDMA passthrough acceleration over a VPC network. By attaching two ERIs to an instance, you can achieve a network bandwidth of 160 Gbit/s if each ERI is connected to a different NIC index. If all ERIs are connected to the same NIC index, the instance can reach up to 100 Gbit/s. For more information, see AttachNetworkInterface.

      Note

      For instructions on how to use ERI, see Enable eRDMA on enterprise-level instances or Enable eRDMA on GPU instances.

The following table lists the instance types and specifications for the ebmgn7ex family.

Instance type

vCPU

Memory (GiB)

GPU memory

Network bandwidth (Gbit/s)

Packet forwarding rate (pps)

Private IPv4s per NIC

IPv6s per NIC

Physical NICs

Multi-queue (primary NIC/secondary NIC)

ENIs

ecs.ebmgn7ex.32xlarge

128

1024

80 GB × 8

160 (80 × 2)

24,000,000

30

30

2

32/32

16

Note

ebmgn7ex instances require an image that supports UEFI boot mode. If you use a custom image, you must set its boot mode property to UEFI. For more information, see Instance boot mode.

GPU-accelerated elastic bare metal instance family: ebmgn7e

  • Overview: The ebmgn7e instance family, built on the Shenlong architecture, delivers high performance and flexibility through software-defined hardware.

  • Scenarios:

    • Various deep learning training and development workloads.

    • High-performance computing (HPC) acceleration and simulation.

    Important

    When you use AI training services that feature a high communication load, such as transformer models, you must enable NVLink for GPU-to-GPU communication. Otherwise, data may be damaged due to unpredictable failures that are caused by large-scale data transmission over Peripheral Component Interconnect Express (PCIe) links. If you do not understand the topology of the communication links that are used for AI training services, submit a ticket to obtain technical support.

  • Compute:

    • Processor: Uses the Intel® Xeon® Scalable platform, featuring a 2.9 GHz base frequency, a 3.5 GHz all-core turbo frequency, and PCIe 4.0 support.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Provides a packet forwarding rate of up to 24 million PPS.

The following table lists the specifications of the ebmgn7e instance family.

Instance type

vCPUs

Memory (GiB)

GPU memory

Baseline bandwidth (Gbit/s)

Network PPS

Multi-queue (primary ENI/secondary ENI)

ENIs

Private IPv4s per ENI

IPv6s per ENI

ecs.ebmgn7e.32xlarge

128

1,024

80 GB × 8

64

24,000,000

32/12

32

10

1

You must check and decide whether to enable or disable the Multi-Instance GPU (MIG) feature after an ebmgn7e instance starts. The system does not guarantee whether the MIG feature is enabled or disabled. For more information about MIG, see the NVIDIA Multi-Instance GPU User Guide.

The following table shows which ebmgn7e instance types support the MIG feature.

Instance type

MIG support

Description

ecs.ebmgn7e.32xlarge

Yes

The ebmgn7e bare metal instance supports the MIG feature.

GPU-accelerated elastic bare metal instance family ebmgn7ix

  • Instance family overview:

    • ebmgn7ix is Alibaba Cloud's new Elastic Bare Metal Server instance family, designed for the growing demands of generative AI workloads. Each instance is a bare metal host equipped with eight NVIDIA A10 GPUs.

    • Powered by the latest CIPU 1.0 cloud processor, these instances decouple compute from storage, allowing you to flexibly select the storage resources you need. This design increases the bandwidth between instances to 160 Gbit/s compared to the previous generation, accelerating data transfer and processing for small-scale multi-machine training workloads.

    • Unlike traditional virtualized instances, these bare metal instances support P2P communication between GPUs, significantly improving multi-GPU computing efficiency.

  • Use cases:

    • Use GRID images from Cloud Marketplace to unlock the A10 GPU's graphics capabilities for high-performance graphics processing in animation, film and television special effects production, and rendering.

    • Use ACK for efficient and cost-effective container management to support AIGC graphics generation and LLM inference on models with up to 130 billion parameters.

    • Other general AI workloads, such as image and speech recognition.

  • Compute:

    • These instances are equipped with NVIDIA A10 GPUs:

      • Based on the innovative Ampere architecture.

      • Supports common acceleration features, including vGPU, RTX, and TensorRT.

    • Processor: Intel® Xeon® Scalable processor (Ice Lake) with a 2.9 GHz base frequency and a 3.5 GHz all-core turbo frequency.

  • Storage:

  • Network:

The following table lists the instance types and specifications of the ebmgn7ix instance family.

Instance type

vCPU

Memory (GiB)

GPU

Network bandwidth (Gbit/s)

Packet rate (PPS)

Private IPv4 addresses/ENI

IPv6 addresses/ENI

Multi-queue

ENIs

ecs.ebmgn7ix.32xlarge

128

512

NVIDIA A10 * 8

160

24,000,000

30

30

32/32

16

Note

You must launch ebmgn7ix instances from images that support UEFI. If you use a custom image, ensure its boot mode is set to UEFI. For more information, see boot mode.

Ebmgn7i: GPU-accelerated bare metal instances

  • Instance family overview: The ebmgn7i instance family is based on the X-Dragon architecture and leverages software-defined hardware computing to provide both elasticity and high performance.

  • Use cases:

    • Equipped with high-performance CPUs, memory, and GPUs, these instances can process numerous concurrent AI inference tasks, making them ideal for image, speech, and behavior recognition.

    • With RTX support and high-frequency CPUs, these instances offer high-performance 3D graphics virtualization, ideal for remote graphic design, cloud gaming, and other demanding graphics workloads.

    • With RTX support and high network and cloud disk bandwidth, these instances are ideal for building high-performance render farms.

    • Equipped with multiple GPUs and high network bandwidth, these instances are ideal for small-scale deep learning training.

  • Compute:

    • Powered by NVIDIA A10 GPUs:

      • Innovative Ampere architecture.

      • Support for common acceleration features such as vGPU, RTX, and TensorRT.

    • Processor: 2.9 GHz Intel® Xeon® Scalable processor (Ice Lake) with an all-core turbo frequency of 3.5 GHz.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • High-performance networking with a packet forwarding rate (PPS) of up to 24,000,000.

The following table describes the instance types and specifications of the ebmgn7i instance family.

Instance type

vCPU

Memory (GiB)

GPU

GPU memory

Baseline bandwidth (Gbit/s)

Packet forwarding rate (PPS)

Queues

ENIs

Private IPv4 addresses

IPv6 addresses

ecs.ebmgn7i.32xlarge

128

768

NVIDIA A10 × 4

24 GB × 4

64

24,000,000

32

32

10

1

GPU compute instance family ebmgn7

  • Instance family overview: The ebmgn7 is an instance family built on the X-Dragon architecture that delivers powerful, flexible, and elastic software-defined computing.

  • Use cases:

    • Deep learning, such as training AI algorithms for image classification, autonomous driving, and speech recognition.

    • GPU-intensive scientific computing, such as computational fluid dynamics (CFD), computational finance, molecular dynamics, and environmental analysis.

  • Compute:

    • Processor: 2.5 GHz Intel® Xeon® Platinum 8269CY (Cascade Lake).

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Network performance scales with the instance type: larger types provide higher performance.

The following table lists the specifications for the ebmgn7 instance family.

Instance type

vCPU

Memory (GiB)

GPU memory

Network bandwidth (Gbit/s)

Packet forwarding rate (PPS)

Multi-queue

ENIs

IPv4 addresses per ENI

IPv6 addresses per ENI

ecs.ebmgn7.26xlarge

104

768

40 GB × 8

30

18,000,000

16

15

10

1

GPU elastic bare metal instance family ebmgn6e

  • Instance family overview:

    • The ebmgn6e instance family is built on the X-Dragon architecture and provides software-defined hardware compute that combines elasticity with high performance.

    • This instance family features NVIDIA V100 (32 GB NVLink) GPUs.

    • The V100 (SXM2) GPUs provide the following features:

      • Innovative Volta architecture.

      • 32 GB HBM2 GPU memory per GPU (900 GB/s GPU memory bandwidth).

      • 5,120 CUDA Cores per GPU.

      • 640 Tensor Cores per GPU.

      • Each GPU supports six bidirectional NVLink connections. Each link provides 25 GB/s of unidirectional bandwidth, for a total bidirectional bandwidth of 300 GB/s.

  • Use cases:

    • Deep learning, including the training and inference of artificial intelligence (AI) algorithms for tasks such as image classification, autonomous driving, and speech recognition.

    • Scientific computing, such as computational fluid dynamics (CFD), computational finance, molecular dynamics, and environmental analysis.

  • Compute:

    • vCPU-to-memory ratio of 1:8.

    • Processor: Intel ® Xeon ® Platinum 8163 (Skylake) with a 2.5 GHz base frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Network performance scales with the instance type.

This table lists the instance types and their specifications for the ebmgn6e instance family.

Instance type

vCPUs

Memory (GiB)

GPU

GPU memory

Network bandwidth (Gbit/s)

Network PPS

Multi-queue

ENIs

Private IPv4 addresses

IPv6 addresses

ecs.ebmgn6e.24xlarge

96

768

NVIDIA V100 * 8

32 GB * 8

32

4,800,000

16

15

10

1

Gpu-accelerated bare metal instance family ebmgn6v

  • Instance family overview:

    • ebmgn6v is an instance family built on the X-Dragon architecture. It offers the performance of bare metal servers with the elasticity and flexibility of the cloud.

    • Equipped with NVIDIA V100 GPUs.

    • The GPU accelerator is the V100 (SXM2 package), with the following specifications:

      • Innovative Volta architecture

      • 16 GB of HBM2 GPU memory per GPU, providing 900 GB/s of memory bandwidth.

      • 5,120 CUDA Cores per GPU

      • 640 Tensor Cores per GPU

      • Each GPU supports six bidirectional NVLink links. Each link provides 25 GB/s of unidirectional bandwidth, resulting in a total bidirectional bandwidth of 300 GB/s (6 × 25 × 2).

  • Use cases:

    • Deep learning applications, including model training and inference for image classification, autonomous driving, and speech recognition.

    • Scientific computing, such as computational fluid dynamics (CFD), computational finance, molecular dynamics, and environmental analysis.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: 2.5 GHz Intel ® Xeon ® Platinum 8163 (Skylake).

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Network performance scales with the instance type.

The following table lists the instance types in the ebmgn6v instance family and their corresponding performance metrics.

Instance type

vCPU

Memory (GiB)

GPU

GPU memory

Baseline bandwidth (Gbit/s)

Network PPS

Queues

ENIs

Private IPv4 addresses

IPv6 addresses

ecs.ebmgn6v.24xlarge

96

384

NVIDIA V100 * 8

16 GB * 8

30

4,500,000

8

32

10

1

GPU compute-optimized Elastic Bare Metal Server instance family ebmgn6i

  • Introduction:

    • The ebmgn6i instance family is built on the X-Dragon architecture. These instances use software-defined hardware computing to deliver high performance, flexibility, and elasticity.

    • GPU accelerators: NVIDIA T4:

      • Innovative Turing architecture

      • 16 GB of GPU memory per GPU (320 GB/s of memory bandwidth)

      • 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

  • Scenarios:

    • AI (DL/ML) inference, 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 rendering for AR/VR.

    • Graphics-intensive computing or graphics workstations.

    • GPU-accelerated databases.

    • High-performance computing (HPC).

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: 2.5 GHz Intel® Xeon® Platinum 8163 (Skylake).

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Network performance scales with the instance type.

The following table lists the specifications for the ebmgn6i instance family.

Instance type

vCPUs

Memory (GiB)

GPU

GPU memory

Baseline bandwidth (Gbit/s)

Network PPS

Queues

ENIs

Private IPv4s per ENI

IPv6s per ENI

ecs.ebmgn6i.24xlarge

96

384

NVIDIA T4 × 4

16 GB × 4

30

4,500,000

8

32

10

1

ebmg5s: Network-enhanced elastic bare metal server instance family

  • Instance family overview: Offers dedicated hardware resources and physical isolation.

  • Scenarios:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Containers, and Pouch.

    • Intensive database applications for medium- to large-sized enterprises.

    • Video encoding.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: Intel® Xeon® Platinum 8163 (Skylake) processor with a 2.5 GHz base frequency and a 2.7 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers 4,500,000 network PPS.

The following table lists the instance types and specifications of the ebmg5s instance family.

Instance type

vCPUs

Memory (GiB)

Network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4s per ENI

IPv6s per ENI

Baseline IOPS

Disk bandwidth (Gbit/s)

ecs.ebmg5s.24xlarge

96

384

32

4,500,000

1,800,000

32

10

1

200,000

16

General-purpose ebmg5 instance family

  • Instance family overview: Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Supports third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers such as Docker, Clear Container, and Pouch.

    • Demanding database applications for medium-to-large enterprises.

    • Video encoding.

  • Compute:

    • vCPU-to-memory ratio of 1:4.

    • Processor: 2.5 GHz Intel® Xeon® Platinum 8163 (Skylake) with an all-core turbo frequency of 2.7 GHz.

  • Storage:

    • These are I/O optimized instances.

    • Supported cloud disk types: SSD cloud disk and ESSD. For more information, see Block storage overview.

  • Network:

    • Supports only IPv4.

    • Provides a network PPS of 4,500,000.

The following table lists the specifications of the ebmg5 instance family.

Instance type

vCPU

Memory (GiB)

Network baseline bandwidth (Gbit/s)

Network PPS

Elastic network interfaces

Private IPv4 addresses per ENI

ecs.ebmg5.24xlarge

96

384

10

4,500,000

32

10

Compute-optimized and network-enhanced ECS Bare Metal instances ebmc5s

  • Instance family overview: Provides dedicated hardware resources and physical isolation.

  • Scenarios:

    • High packet-rate scenarios, such as live video comments and telecom service forwarding.

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatible with third-party hypervisors to support hybrid cloud and multi-cloud deployment.

    • Containers (including Docker, Clear Container, and Pouch).

    • Video encoding, decoding, and rendering.

  • Compute:

    • vCPU-to-memory ratio of 1:2.

    • Processor: Intel® Xeon® Platinum 8163 (Skylake) with a 2.5 GHz base frequency and a 2.7 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • These instances support IPv4 and IPv6. For information about IPv6 communication, see IPv6 communication.

    • Delivers a packet forwarding rate of 4.5 million PPS.

This table lists the instance types and metrics for the ebmc5s instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

Elastic network interfaces

Private IPv4 addresses

IPv6 addresses

Baseline IOPS

Baseline disk bandwidth (Gbit/s)

ecs.ebmc5s.24xlarge

96

192

32

4,500,000

1,800,000

32

10

1

200,000

16

ebmr5s instance family: Memory-optimized and network-enhanced

  • Instance family overview: Provides dedicated hardware resources and physical isolation.

  • Use cases:

    • Workloads that require direct access to physical resources or hardware-bound licenses.

    • Compatible with third-party hypervisors for hybrid cloud and multi-cloud deployments.

    • Containers, such as Docker, Clear Container, and Pouch.

    • High-performance databases and in-memory databases.

    • Data analytics and mining, and distributed in-memory caches.

    • Hadoop and Spark clusters, and other memory-intensive enterprise applications.

  • Compute:

    • vCPU-to-memory ratio of 1:8.

    • Processor: Intel® Xeon® Platinum 8163 (Skylake) with a 2.5 GHz base frequency and a 2.7 GHz all-core turbo frequency.

  • Storage:

  • Network:

    • Supports only IPv4.

    • Achieves up to 4,500,000 network PPS.

The following table lists the instance types and specifications of the ebmr5s instance family.

Instance type

vCPU

Memory (GiB)

Baseline network bandwidth (Gbit/s)

Network PPS

Connections

ENIs

Private IPv4s per ENI

Baseline cloud disk IOPS

Baseline cloud disk bandwidth (Gbit/s)

ecs.ebmr5s.24xlarge

96

768

32

4,500,000

1,800,000

32

10

200,000

16

Billing

Elastic Bare Metal Server supports pay-as-you-go and subscription billing methods. For a comparison, see Billing method overview.