Amazon Web Services (AWS) has announced the general availability of its new Amazon Elastic Compute Cloud (Amazon EC2) G7 instances, marking the first time a major cloud provider has integrated NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs into its infrastructure. Designed to support intensive AI inference, graphics rendering, and data analytics, these instances are now accessible to users in the US East (Ohio) and US West (Oregon) regions, according to an official AWS compute blog post.
The G7 instances represent a significant architectural shift from the previous G6 generation, utilizing custom sixth-generation Intel Xeon Scalable processors to drive performance gains. AWS reports that these instances deliver up to 4.6 times the performance for AI inference workloads and up to 2.1 times the performance for graphics tasks compared to their predecessors. This infrastructure is intended to assist organizations that require high-bandwidth connectivity for spatial computing, video transcoding, and virtual desktop infrastructure (VDI), as detailed in the official Amazon EC2 G7 product page.
Technical Specifications and Architectural Upgrades
The core of the G7 performance improvement lies in the hardware configuration of the NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs. Each GPU is equipped with 32 GB of memory, which provides a 1.33-fold increase in capacity and a 2.45-fold increase in memory bandwidth compared to the G6 instances. The architecture also incorporates 5th Gen Tensor Cores and 4th Gen RT Cores, which are specifically optimized for accelerating AI-driven inference and complex ray-tracing graphics workflows, according to AWS service update documentation.
Networking and storage capabilities have also seen substantial upgrades. The G7 instances feature 700 Gbps of networking throughput enabled by Elastic Fabric Adapter (EFA), representing a seven-fold increase over G6 networking speeds. To minimize data latency, the instances support up to 7.6 TB of local NVMe SSD storage, allowing large datasets and machine learning models to remain in close proximity to the compute resources. Furthermore, the inclusion of ninth-generation NVENC and sixth-generation NVDEC engines provides support for 4:2:2 video encoding and decoding, enabling 1.5 times more concurrent video streams than the previous generation.
Instance Configuration Options
AWS offers the G7 instances in seven distinct sizes to accommodate varying workload requirements. The largest standard configuration, the g7.48xlarge, includes 8 NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, 192 vCPUs, and 768 GiB of system memory. Below is the breakdown of the instance specifications as provided by official Amazon service documentation:
| Instance Name | GPUs | vCPUs | Memory (GiB) | Network Bandwidth (Gbps) |
|---|---|---|---|---|
| g7.2xlarge | 1 | 8 | 32 | Up to 60 |
| g7.4xlarge | 1 | 16 | 64 | Up to 100 |
| g7.8xlarge | 1 | 32 | 128 | Up to 100 |
| g7.12xlarge | 2 | 48 | 192 | 175 |
| g7.24xlarge | 4 | 96 | 384 | 350 |
| g7.48xlarge | 8 | 192 | 768 | 700 |
| g7.metal | 8 | 192 | 768 | 700 |
Software Compatibility and Deployment
To facilitate the transition to G7, AWS provides support for multiple operating systems, including Amazon Linux, Ubuntu, RHEL, and Windows Server. Developers can utilize AWS Deep Learning AMIs (DLAMI) or NVIDIA Workstation AMIs, which come with pre-packaged drivers. For those working within containerized environments, the instances support Amazon EKS when using NVIDIA driver version R595, as noted in the official AWS EC2 user guide.
The instances also support advanced NVIDIA technologies, including GPUDirect P2P for multi-GPU communication and GPUDirect RDMA with EFA for low-latency data transfers. These features are designed to enhance performance for distributed multi-node workloads, particularly when paired with Amazon FSx for Lustre. Users interested in exploring these instances can access them through On-Demand, Savings Plans, or Spot Instance purchasing options via the Amazon EC2 console. Future regional expansion plans for these instances are tracked through the AWS Capabilities by Region resource page.
Feedback regarding the performance and deployment of G7 instances can be directed to the AWS support team or shared on the AWS re:Post community forum, which serves as the primary channel for technical community engagement regarding these updates.