Executive Summary
Amazon Web Services (AWS) has announced the general availability of its new Amazon EC2 G7e instances, powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. These instances are designed to provide cost-effective, high-performance computing for generative AI inference, graphics-intensive applications, and scientific computing. Compared to the previous G6e generation, G7e instances offer significant improvements in performance, GPU memory, and networking bandwidth, enabling customers to run larger models and more demanding workloads.
Key Takeaways
* Core Technology: G7e instances are accelerated by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.
* Performance Boost: Delivers up to 2.3 times better AI inference performance compared to the preceding G6e instances.
* Enhanced Memory: Offers double the GPU memory (96 GB per GPU) and 1.85 times the memory bandwidth of G6e, allowing medium-sized models (up to 70B parameters) to run on a single GPU.
* Advanced Networking: Features four times the networking bandwidth (up to 1,600 Gbps) and supports NVIDIA GPUDirect RDMA with Elastic Fabric Adapter (EFA) for efficient multi-node workloads.
* Multi-GPU Scaling: Supports NVIDIA GPUDirect P2P for low-latency communication between GPUs within an instance, with configurations scaling up to 8 GPUs and 768 GB of total GPU memory.
* Availability: Immediately available in US East (N. Virginia) and US East (Ohio) AWS Regions, with purchasing options including On-Demand, Spot Instances, and Savings Plans.
Strategic Importance
This launch reinforces AWS's commitment to providing state-of-the-art compute for the rapidly growing AI market, specifically targeting the cost-sensitive but performance-critical area of AI model inference. By integrating the latest NVIDIA Blackwell architecture, AWS remains competitive in attracting and retaining customers with demanding AI and graphics workloads.