EFA basics Supported interfaces and libraries Supported instance types Supported operating systems EFA limitations EFA pricing

Elastic Fabric Adapter for AI/ML and HPC workloads on Amazon EC2

An Elastic Fabric Adapter (EFA) is a network device that you can attach to your Amazon EC2 instance to accelerate Artificial Intelligence (AI), Machine Learning (ML), and High Performance Computing (HPC) applications. EFA enables you to achieve the application performance of an on-premises AI/ML or HPC cluster, with the scalability, flexibility, and elasticity provided by the AWS Cloud.

EFA provides lower and more consistent latency and higher throughput than the TCP transport traditionally used in cloud-based HPC systems. It enhances the performance of inter-instance communication that is critical for scaling AI/ML and HPC applications. It is optimized to work on the existing AWS network infrastructure and it can scale depending on application requirements.

EFA integrates with Libfabric, and it supports NVIDIA Collective Communications Library (NCCL), NVIDIA Inference Xfer Library (NIXL), and Expert Parallelism (EP) library for AI and ML applications, and Open MPI 4.1 and later and Intel MPI 2019 Update 5 and later for HPC applications. NCCL and MPI integrate with Libfabric 1.7.0 and later. NIXL integrates with Libfabric 1.21.0 and later.

EFA supports RDMA (Remote Direct Memory Access) write on most supported instance types that have Nitro version 4 and later. RDMA read is supported on all instances with Nitro version 4 and later. GPUDirect RDMA (GDR) and GPUDirect Async (GDA) are available on select Nitro v4 and later instances. For more information, see Supported instance types.

EFA basics

An EFA device can be attached to an EC2 instance in two ways:

Using a traditional EFA interface, also called EFA with ENA, which creates both an EFA device and an ENA device.
Using an EFA-only interface, which creates just the EFA device.

The EFA device provides capabilities like built-in OS-bypass and congestion control through the Scalable Reliable Datagram (SRD) protocol. The EFA device features enable low-latency, reliable transport functionality that allows EFA interface to provide better application performance for HPC and ML applications on Amazon EC2. While the ENA device offers traditional IP networking. For guidance on which instance configuration to use based on your use case, see Maximize network bandwidth.

Contrasting a traditional HPC software stack with one that uses an EFA.

Traditionally, AI/ML applications use NCCL and NIXL (for disaggregated inference). HPC applications use the Message Passing Interface (MPI) to interface with the system's network transport. In the AWS cloud, this has meant that applications interface with NCCL, NIXL, or MPI, which then uses the operating system's TCP/IP stack and the ENA device driver to enable network communication between instances.

With a traditional EFA (EFA with ENA) or EFA-only interface, AI/ML applications use NCCL and NIXL (for disaggregated inference). HPC applications use MPI, to interface directly with the Libfabric API. The Libfabric API bypasses the operating system kernel and communicates directly with the EFA device to put packets on the network. This reduces overhead and enables AI/ML and HPC applications to run more efficiently.

Note

Libfabric is a core component of the OpenFabrics Interfaces (OFI) framework, which defines and exports the user-space API of OFI. For more information, see the Libfabric OpenFabrics website.

Differences between ENA, EFA, and EFA-only network interfaces

Amazon EC2 provides two types of network interfaces:

ENA interfaces provide all of the traditional IP networking and routing features that are required to support IP networking for a VPC. For more information, see Enable enhanced networking with ENA on your EC2 instances.
EFA (EFA with ENA) interfaces provide both the ENA device for IP networking and the EFA device for low-latency, high-throughput communication.
EFA-only interfaces support only the EFA device capabilities, without the ENA device for traditional IP networking.

The following table provides a comparison of ENA, EFA, and EFA-only network interfaces.

	ENA	EFA (EFA with ENA)	EFA-only
Supports IP networking functionality	Yes	Yes	No
Can be assigned IPv4 or IPv6 addresses	Yes	Yes	No
Can be used as primary network interface for instance	Yes	Yes	No
Counts towards ENI attachment limit for instance	Yes	Yes	Yes
Instance type support	Supported on all Nitro-based instances types	Supported instance types	Supported instance types
Parameter naming in EC2 APIs	`interface`	`efa`	`efa-only`
Field naming in EC2 console	No selection	EFA with ENA	EFA-only

Supported interfaces and libraries

EFAs support the following interfaces and libraries:

Open MPI 4.1 and later
Intel MPI 2019 Update 5 and later
NVIDIA Collective Communications Library (NCCL) 2.4.2 and later
NVIDIA Inference Xfer Library (NIXL) 1.0.0 and later
DeepEP v2
AWS Neuron SDK version 2.3 and later

Supported instance types

All of the following instance types support EFA. Additionally, the tables indicate RDMA read, RDMA write, GPUDirect RDMA (GDR), and GPUDirect Async (GDA) support for the instance types.

Nitro v6 (EFA v4)

Instance type	RDMA read support	RDMA write support	GPUDirect RDMA (GDR)	GPUDirect Async (GDA)
General Purpose
m8a.48xlarge	Yes	Yes	No	No
m8a.metal-48xl	Yes	Yes	No	No
m8azn.24xlarge	Yes	Yes	No	No
m8azn.metal-24xl	Yes	Yes	No	No
m8gb.16xlarge	Yes	Yes	No	No
m8gb.24xlarge	Yes	Yes	No	No
m8gb.48xlarge	Yes	Yes	No	No
m8gb.metal-24xl	Yes	Yes	No	No
m8gb.metal-48xl	Yes	Yes	No	No
m8gn.16xlarge	Yes	Yes	No	No
m8gn.24xlarge	Yes	Yes	No	No
m8gn.48xlarge	Yes	Yes	No	No
m8gn.metal-24xl	Yes	Yes	No	No
m8gn.metal-48xl	Yes	Yes	No	No
m8i.48xlarge	Yes	Yes	No	No
m8i.96xlarge	Yes	Yes	No	No
m8i.metal-48xl	Yes	Yes	No	No
m8i.metal-96xl	Yes	Yes	No	No
m8id.48xlarge	Yes	Yes	No	No
m8id.96xlarge	Yes	Yes	No	No
m8id.metal-48xl	Yes	Yes	No	No
m8id.metal-96xl	Yes	Yes	No	No
m8in.48xlarge	Yes	Yes	No	No
m8in.96xlarge	Yes	Yes	No	No
m8in.metal-48xl	Yes	Yes	No	No
m8in.metal-96xl	Yes	Yes	No	No
m8idn.48xlarge	Yes	Yes	No	No
m8idn.96xlarge	Yes	Yes	No	No
m8idn.metal-48xl	Yes	Yes	No	No
m8idn.metal-96xl	Yes	Yes	No	No
m8ib.48xlarge	Yes	Yes	No	No
m8ib.96xlarge	Yes	Yes	No	No
m8ib.metal-48xl	Yes	Yes	No	No
m8ib.metal-96xl	Yes	Yes	No	No
m8idb.48xlarge	Yes	Yes	No	No
m8idb.96xlarge	Yes	Yes	No	No
m8idb.metal-48xl	Yes	Yes	No	No
m8idb.metal-96xl	Yes	Yes	No	No
m9g.48xlarge	Yes	Yes	No	No
m9g.metal-48xl	Yes	Yes	No	No
m9gd.48xlarge	Yes	Yes	No	No
m9gd.metal-48xl	Yes	Yes	No	No
Compute Optimized
c8a.48xlarge	Yes	Yes	No	No
c8a.metal-48xl	Yes	Yes	No	No
c8gb.16xlarge	Yes	Yes	No	No
c8gb.24xlarge	Yes	Yes	No	No
c8gb.48xlarge	Yes	Yes	No	No
c8gb.metal-24xl	Yes	Yes	No	No
c8gb.metal-48xl	Yes	Yes	No	No
c8gn.16xlarge	Yes	Yes	No	No
c8gn.24xlarge	Yes	Yes	No	No
c8gn.48xlarge	Yes	Yes	No	No
c8gn.metal-24xl	Yes	Yes	No	No
c8gn.metal-48xl	Yes	Yes	No	No
c8i.48xlarge	Yes	Yes	No	No
c8i.96xlarge	Yes	Yes	No	No
c8i.metal-48xl	Yes	Yes	No	No
c8i.metal-96xl	Yes	Yes	No	No
c8id.48xlarge	Yes	Yes	No	No
c8id.96xlarge	Yes	Yes	No	No
c8id.metal-48xl	Yes	Yes	No	No
c8id.metal-96xl	Yes	Yes	No	No
c8in.48xlarge	Yes	Yes	No	No
c8in.96xlarge	Yes	Yes	No	No
c8in.metal-48xl	Yes	Yes	No	No
c8in.metal-96xl	Yes	Yes	No	No
c8ib.48xlarge	Yes	Yes	No	No
c8ib.96xlarge	Yes	Yes	No	No
c8ib.metal-48xl	Yes	Yes	No	No
c8ib.metal-96xl	Yes	Yes	No	No
c9g.48xlarge	Yes	Yes	No	No
c9g.metal-48xl	Yes	Yes	No	No
c9gd.48xlarge	Yes	Yes	No	No
c9gd.metal-48xl	Yes	Yes	No	No
Memory Optimized
r8a.48xlarge	Yes	Yes	No	No
r8a.metal-48xl	Yes	Yes	No	No
r8gb.16xlarge	Yes	Yes	No	No
r8gb.24xlarge	Yes	Yes	No	No
r8gb.48xlarge	Yes	Yes	No	No
r8gb.metal-24xl	Yes	Yes	No	No
r8gb.metal-48xl	Yes	Yes	No	No
r8gn.16xlarge	Yes	Yes	No	No
r8gn.24xlarge	Yes	Yes	No	No
r8gn.48xlarge	Yes	Yes	No	No
r8gn.metal-24xl	Yes	Yes	No	No
r8gn.metal-48xl	Yes	Yes	No	No
r8i.48xlarge	Yes	Yes	No	No
r8i.96xlarge	Yes	Yes	No	No
r8i.metal-48xl	Yes	Yes	No	No
r8i.metal-96xl	Yes	Yes	No	No
r8id.48xlarge	Yes	Yes	No	No
r8id.96xlarge	Yes	Yes	No	No
r8id.metal-48xl	Yes	Yes	No	No
r8id.metal-96xl	Yes	Yes	No	No
r8in.48xlarge	Yes	Yes	No	No
r8in.96xlarge	Yes	Yes	No	No
r8in.metal-48xl	Yes	Yes	No	No
r8in.metal-96xl	Yes	Yes	No	No
r8idn.48xlarge	Yes	Yes	No	No
r8idn.96xlarge	Yes	Yes	No	No
r8idn.metal-48xl	Yes	Yes	No	No
r8idn.metal-96xl	Yes	Yes	No	No
r8ib.48xlarge	Yes	Yes	No	No
r8ib.96xlarge	Yes	Yes	No	No
r8ib.metal-48xl	Yes	Yes	No	No
r8ib.metal-96xl	Yes	Yes	No	No
r8idb.48xlarge	Yes	Yes	No	No
r8idb.96xlarge	Yes	Yes	No	No
r8idb.metal-48xl	Yes	Yes	No	No
r8idb.metal-96xl	Yes	Yes	No	No
r9g.48xlarge	Yes	Yes
r9g.metal-48xl	Yes	Yes
r9gd.48xlarge	Yes	Yes
r9gd.metal-48xl	Yes	Yes
x8aedz.24xlarge	Yes	Yes	No	No
x8aedz.metal-24xl	Yes	Yes	No	No
x8i.48xlarge	Yes	Yes	No	No
x8i.64xlarge	Yes	Yes	No	No
x8i.96xlarge	Yes	Yes	No	No
x8i.metal-48xl	Yes	Yes	No	No
x8i.metal-96xl	Yes	Yes	No	No
Storage Optimized
i8ge.48xlarge	Yes	Yes	No	No
i8ge.metal-48xl	Yes	Yes	No	No
Accelerated Computing
g7.8xlarge	Yes	Yes	Yes	Yes
g7.12xlarge	Yes	Yes	Yes	Yes
g7.24xlarge	Yes	Yes	Yes	Yes
g7.48xlarge	Yes	Yes	Yes	Yes
g7e.8xlarge	Yes	Yes	No	No
g7e.12xlarge	Yes	Yes	Yes	Yes
g7e.24xlarge	Yes	Yes	Yes	Yes
g7e.48xlarge	Yes	Yes	Yes	Yes
p6e-gb200.36xlarge	Yes	Yes	Yes	No
p6-b200.48xlarge	Yes	Yes	Yes	Yes
p6-b300.48xlarge	Yes	Yes	Yes	Yes
High Performance Computing
hpc8a.96xlarge	Yes	Yes	No	No

Nitro v5 (EFA v3)

Instance type	RDMA read support	RDMA write support	GPUDirect RDMA (GDR)	GPUDirect Async (GDA)
General Purpose
m8g.24xlarge	Yes	Yes	No	No
m8g.48xlarge	Yes	Yes	No	No
m8g.metal-24xl	Yes	Yes	No	No
m8g.metal-48xl	Yes	Yes	No	No
m8gd.24xlarge	Yes	Yes	No	No
m8gd.48xlarge	Yes	Yes	No	No
m8gd.metal-24xl	Yes	Yes	No	No
m8gd.metal-48xl	Yes	Yes	No	No
Compute Optimized
c7gn.16xlarge	Yes	No	No	No
c7gn.metal	Yes	No	No	No
c8g.24xlarge	Yes	Yes	No	No
c8g.48xlarge	Yes	Yes	No	No
c8g.metal-24xl	Yes	Yes	No	No
c8g.metal-48xl	Yes	Yes	No	No
c8gd.24xlarge	Yes	Yes	No	No
c8gd.48xlarge	Yes	Yes	No	No
c8gd.metal-24xl	Yes	Yes	No	No
c8gd.metal-48xl	Yes	Yes	No	No
Memory Optimized
r8g.24xlarge	Yes	Yes	No	No
r8g.48xlarge	Yes	Yes	No	No
r8g.metal-24xl	Yes	Yes	No	No
r8g.metal-48xl	Yes	Yes	No	No
r8gd.24xlarge	Yes	Yes	No	No
r8gd.48xlarge	Yes	Yes	No	No
r8gd.metal-24xl	Yes	Yes	No	No
r8gd.metal-48xl	Yes	Yes	No	No
x8g.24xlarge	Yes	Yes	No	No
x8g.48xlarge	Yes	Yes	No	No
x8g.metal-24xl	Yes	Yes	No	No
x8g.metal-48xl	Yes	Yes	No	No
Storage Optimized
i7ie.48xlarge	Yes	Yes	No	No
i7ie.metal-48xl	Yes	Yes	No	No
i8g.48xlarge	Yes	Yes	No	No
i8g.metal-48xl	Yes	Yes	No	No
Accelerated Computing
p5en.48xlarge	Yes	Yes	Yes	Yes
trn2.3xlarge	Yes	Yes	No	No
trn2.48xlarge	Yes	Yes	No	No
trn2u.48xlarge	Yes	Yes	No	No
High Performance Computing
hpc7g.4xlarge	Yes	No	No	No
hpc7g.8xlarge	Yes	No	No	No
hpc7g.16xlarge	Yes	No	No	No

Nitro v4 (EFA v2)

Instance type	RDMA read support	RDMA write support	GPUDirect RDMA (GDR)	GPUDirect Async (GDA)
General Purpose
m6a.48xlarge	Yes	Yes	No	No
m6a.metal	Yes	Yes	No	No
m6i.32xlarge	Yes	Yes	No	No
m6i.metal	Yes	Yes	No	No
m6id.32xlarge	Yes	Yes	No	No
m6id.metal	Yes	Yes	No	No
m6idn.32xlarge	Yes	Yes	No	No
m6idn.metal	Yes	Yes	No	No
m6in.32xlarge	Yes	Yes	No	No
m6in.metal	Yes	Yes	No	No
m7a.48xlarge	Yes	Yes	No	No
m7a.metal-48xl	Yes	Yes	No	No
m7g.16xlarge	Yes	Yes	No	No
m7g.metal	Yes	Yes	No	No
m7gd.16xlarge	Yes	Yes	No	No
m7gd.metal	Yes	Yes	No	No
m7i.48xlarge	Yes	Yes	No	No
m7i.metal-48xl	Yes	Yes	No	No
Compute Optimized
c6a.48xlarge	Yes	Yes	No	No
c6a.metal	Yes	Yes	No	No
c6gn.16xlarge	Yes	Yes	No	No
c6i.32xlarge	Yes	Yes	No	No
c6i.metal	Yes	Yes	No	No
c6id.32xlarge	Yes	Yes	No	No
c6id.metal	Yes	Yes	No	No
c6in.32xlarge	Yes	Yes	No	No
c6in.metal	Yes	Yes	No	No
c7a.48xlarge	Yes	Yes	No	No
c7a.metal-48xl	Yes	Yes	No	No
c7g.16xlarge	Yes	Yes	No	No
c7g.metal	Yes	Yes	No	No
c7gd.16xlarge	Yes	Yes	No	No
c7gd.metal	Yes	Yes	No	No
c7i.48xlarge	Yes	Yes	No	No
c7i.metal-48xl	Yes	Yes	No	No
Memory Optimized
r6a.48xlarge	Yes	Yes	No	No
r6a.metal	Yes	Yes	No	No
r6i.32xlarge	Yes	Yes	No	No
r6i.metal	Yes	Yes	No	No
r6id.32xlarge	Yes	Yes	No	No
r6id.metal	Yes	Yes	No	No
r6idn.32xlarge	Yes	Yes	No	No
r6idn.metal	Yes	Yes	No	No
r6in.32xlarge	Yes	Yes	No	No
r6in.metal	Yes	Yes	No	No
r7a.48xlarge	Yes	Yes	No	No
r7a.metal-48xl	Yes	Yes	No	No
r7g.16xlarge	Yes	Yes	No	No
r7g.metal	Yes	Yes	No	No
r7gd.16xlarge	Yes	Yes	No	No
r7gd.metal	Yes	Yes	No	No
r7i.48xlarge	Yes	Yes	No	No
r7i.metal-48xl	Yes	Yes	No	No
r7iz.32xlarge	Yes	Yes	No	No
r7iz.metal-32xl	Yes	Yes	No	No
u7i-6tb.112xlarge	Yes	Yes	No	No
u7i-8tb.112xlarge	Yes	Yes	No	No
u7i-12tb.224xlarge	Yes	Yes	No	No
u7in-16tb.224xlarge	Yes	Yes	No	No
u7in-24tb.224xlarge	Yes	Yes	No	No
u7in-32tb.224xlarge	Yes	Yes	No	No
u7inh-32tb.480xlarge	Yes	Yes	No	No
x2idn.32xlarge	Yes	Yes	No	No
x2idn.metal	Yes	Yes	No	No
x2iedn.32xlarge	Yes	Yes	No	No
x2iedn.metal	Yes	Yes	No	No
Storage Optimized
i4g.16xlarge	Yes	Yes	No	No
i4i.32xlarge	Yes	Yes	No	No
i4i.metal	Yes	Yes	No	No
i7i.24xlarge	Yes	Yes	No	No
i7i.48xlarge	Yes	Yes	No	No
i7i.metal-48xl	Yes	Yes	No	No
im4gn.16xlarge	Yes	Yes	No	No
Accelerated Computing
f2.48xlarge	Yes	Yes	No	No
g6.8xlarge	Yes	Yes	No	No
g6.12xlarge	Yes	Yes	No	No
g6.16xlarge	Yes	Yes	No	No
g6.24xlarge	Yes	Yes	No	No
g6.48xlarge	Yes	Yes	No	No
g6e.8xlarge	Yes	Yes	No	No
g6e.12xlarge	Yes	Yes	No	No
g6e.16xlarge	Yes	Yes	No	No
g6e.24xlarge	Yes	Yes	No	No
g6e.48xlarge	Yes	Yes	No	No
gr6.8xlarge	Yes	Yes	No	No
p5.4xlarge	Yes	Yes	No	No
p5.48xlarge	Yes	Yes	Yes	No
p5e.48xlarge	Yes	Yes	Yes	No
trn1.32xlarge	Yes	Yes	No	No
trn1n.32xlarge	Yes	Yes	No	No
High Performance Computing
hpc6a.48xlarge	Yes	Yes	No	No
hpc6id.32xlarge	Yes	Yes	No	No
hpc7a.12xlarge	Yes	Yes	No	No
hpc7a.24xlarge	Yes	Yes	No	No
hpc7a.48xlarge	Yes	Yes	No	No
hpc7a.96xlarge	Yes	Yes	No	No

Nitro v3 (EFA v1)

Instance type	RDMA read support	RDMA write support	GPUDirect RDMA (GDR)	GPUDirect Async (GDA)
General Purpose
m5dn.24xlarge	No	No	No	No
m5dn.metal	No	No	No	No
m5n.24xlarge	No	No	No	No
m5n.metal	No	No	No	No
m5zn.12xlarge	No	No	No	No
m5zn.metal	No	No	No	No
Compute Optimized
c5n.9xlarge	No	No	No	No
c5n.18xlarge	No	No	No	No
c5n.metal	No	No	No	No
Memory Optimized
r5dn.24xlarge	No	No	No	No
r5dn.metal	No	No	No	No
r5n.24xlarge	No	No	No	No
r5n.metal	No	No	No	No
x2iezn.12xlarge	No	No	No	No
x2iezn.metal	No	No	No	No
Storage Optimized
i3en.12xlarge	No	No	No	No
i3en.24xlarge	No	No	No	No
i3en.metal	No	No	No	No
Accelerated Computing
dl2q.24xlarge	No	No	No	No
g4dn.8xlarge	No	No	No	No
g4dn.12xlarge	No	No	No	No
g4dn.16xlarge	No	No	No	No
g4dn.metal	No	No	No	No
g5.8xlarge	No	No	No	No
g5.12xlarge	No	No	No	No
g5.16xlarge	No	No	No	No
g5.24xlarge	No	No	No	No
g5.48xlarge	No	No	No	No
inf1.24xlarge	No	No	No	No
p3dn.24xlarge	No	No	No	No
p4d.24xlarge	Yes	No	No	No
p4de.24xlarge	Yes	No	No	No
vt1.24xlarge	No	No	No	No
Previous Generation
p3dn.24xlarge	No	No	No	No

To see the available instance types that support EFAs in a specific Region

The available instance types vary by Region. To see the available instance types that support EFAs in a Region, use the describe-instance-types command with the --region parameter. Include the --filters parameter to scope the results to the instance types that support EFA and the --query parameter to scope the output to the value of InstanceType.


aws ec2 describe-instance-types \
    --region us-east-1  \
    --filters Name=network-info.efa-supported,Values=true \
    --query "InstanceTypes[*].[InstanceType]"  \
    --output text | sort

Supported operating systems

Operating system support differs depending on the processor type. The following table shows the supported operating systems.

Operating system	Intel/AMD (`x86_64`) instance types	AWS Graviton (`arm64`) instance types
Amazon Linux 2023	✓	✓
RHEL 8, 9, and 10	✓	✓
Debian 11, 12, and 13	✓	✓
Rocky Linux 8 and 9	✓	✓
Ubuntu 22.04, 24.04, and 26.04	✓	✓
SUSE Linux Enterprise 15 SP2 and later	✓	✓

Note

Some of the listed operating systems might not be supported with Intel MPI. If you are using Intel MPI, refer to the Intel MPI documentation to verify support for your operating system.

EFA limitations

EFAs have the following limitations:

RDMA write is not supported with all instance types. For more information, see Supported instance types.
EFA traffic¹ between P4d/P4de/DL1 instances and other instance types is currently not supported.
Instance types that support multiple network cards can be configured with one EFA per network card. All other supported instance types support only one EFA per instance.
c7g.16xlarge, m7g.16xlarge, and r7g.16xlarge Dedicated Instances and Dedicated Hosts are not supported when an EFA is attached.
EFA traffic¹ can't cross Availability Zones or VPCs. This does not apply to normal IP traffic from the ENA device of an EFA interface.
EFA traffic¹ is not routable. Normal IP traffic from the ENA device of an EFA interface remains routable.
EFA is not supported on AWS Outposts.
The EFA device of an EFA (EFA with ENA) interface is supported on Windows instances only for AWS Cloud Digital Interface Software Development Kit (AWS CDI SDK) based applications. If you attach an EFA (EFA with ENA) interface to a Windows instance for non-CDI SDK based applications, it functions as an ENA interface, without the added EFA device capabilities. The EFA-only interface is not supported by AWS CDI based applications on Windows or Linux. For more information, see the AWS Cloud Digital Interface Software Development Kit (AWS CDI SDK) User Guide.

¹EFA traffic refers to the traffic transmitted through the EFA device of either an EFA (EFA with ENA) or EFA-only interface.

EFA pricing

EFA is available as an optional Amazon EC2 networking feature that you can enable on any supported instance at no additional cost.

Warning Javascript is disabled or is unavailable in your browser.

To use the Amazon Web Services Documentation, Javascript must be enabled. Please refer to your browser's Help pages for instructions.

Document Conventions

Optimize network performance on Windows

Get started with EFA and MPI

Elastic Fabric Adapter for AI/ML and HPC workloads on Amazon EC2

Contents

EFA basics

Note

Differences between ENA, EFA, and EFA-only network interfaces

Supported interfaces and libraries

Supported instance types

To see the available instance types that support EFAs in a specific Region

Supported operating systems

Note

EFA limitations

EFA pricing