Announcing Amazon EC2 G7e instances accelerated by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs

• Amazon EC2 G7e instances launched, powered by NVIDIA RTX PRO 6000 Blackwell GPUs. • Deliver up to 2.3× inference performance over G6e, ideal for generative AI and graphics workloads. • Double GPU memory and 1.85× memory bandwidth, enabling 70B‑parameter models in FP8 on single GPU. • GPUDirect P2P and RDMA reduce multi‑GPU latency, quadruple inter‑GPU bandwidth vs L40s. • 4× networking bandwidth supports small‑scale multi‑node workloads and EFA integration. • Supports spatial, scientific, and other GPU‑enabled workloads with cost‑effective scaling.

Article Summaries:

• Amazon announced the general availability of its new EC2 G7e instances, powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. The instances deliver up to 2.3× inference performance over the previous G6e line, with twice the GPU memory and 1.85× memory bandwidth per GPU. They support up to eight GPUs per node, offering 768 GB of total GPU memory and 1,600 Gbps of network bandwidth. Enhanced GPUDirect P2P and RDMA enable low‑latency multi‑GPU and multi‑node workloads, while GPUDirectStorage boosts model‑loading speeds to 1.2 Tbps. The new instances target generative‑AI inference, graphics, spatial and scientific computing.

Sources:

• https://aws.amazon.com/blogs/aws/announcing-amazon-ec2-g7e-instances-accelerated-by-nvidia-rtx-pro-6000-blackwell-server-edition-gpus/