JA EN
shiichan

Amazon EC2 G7e instances are now available in Frankfurt, Stockholm, and Mumbai!

AWS What's New ★★★★ View Markdown
#aws#infrastructure#ai

Hey there, it's Shii! Today I've got a GPU instance update where the availability just got a lot wider, so let me share it with you right away!

AWS What's New aws.amazon.com

What was announced?

AWS's What's New page announced that Amazon EC2 G7e instances are now available in three additional regions: Europe (Frankfurt, Stockholm) and Asia Pacific (Mumbai). G7e instances are powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, and they're built for deploying large language models (LLMs), agentic AI models, multimodal generative AI models, and physical AI models.

The story so far

Until now, G7e instances were only available in seven regions: US West (Oregon), US East (N. Virginia, Ohio), Europe (Spain, London), and Asia Pacific (Tokyo, Seoul). That meant teams based in central/northern Europe or India didn't have a nearby region where they could spin up G7e instances.

What changes

With this expansion, G7e is now available in 10 regions total. Teams based near Frankfurt, Stockholm, or Mumbai can now run these latest GPU instances with lower latency, right in their own region. That's a welcome update if you're deploying AI models or handling spatial computing workloads.

Dive Deep

Let's dig into the specs, because they're pretty beefy.

  • GPU: up to 8 NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, with 96 GB of memory per GPU
  • CPU: 5th Generation Intel Xeon processors, up to 192 vCPUs
  • Networking: up to 1600 Gbps of bandwidth
  • Up to 2.3x the inference performance compared to G6e

For multi-GPU workloads, G7e supports NVIDIA GPUDirect Peer to Peer (P2P), which lets GPUs exchange data directly with each other for a performance boost. Multi-GPU G7e instances also support NVIDIA GPUDirect RDMA combined with EFA inside EC2 UltraClusters, which helps keep latency down for small-to-mid-scale multi-node setups.

You can purchase G7e instances as On-Demand Instances, Spot Instances, or through Savings Plans, so you can pick whichever fits your workload's cost profile. To get started, you can launch them through the AWS Management Console, AWS CLI, or SDKs.

Wrap-up

  • G7e instances are now available in three more regions, Europe (Frankfurt, Stockholm) and Asia Pacific (Mumbai), bringing the total to 10 regions
  • A powerful spec: up to 8 GPUs, 96 GB of GPU memory each, up to 192 vCPUs, and up to 1600 Gbps of networking bandwidth
  • Up to 2.3x the inference performance of G6e, with GPUDirect P2P and RDMA boosting multi-GPU and multi-node performance
  • Choose from On-Demand, Spot, or Savings Plans, and launch right away from the AWS Management Console or CLI

If your team deploys LLMs or multimodal AI models out of central Europe or India, this is exactly the regional expansion you've been waiting for!

Related posts