AWS Unleashes New Blackwell-Powered Instances for AI and Graphics

Innova Tek Solutions Inc. reports that Amazon Web Services has launched its latest generation of GPU-powered cloud instances. The new Amazon EC2 G7e instances, now generally available, are built around NVIDIA's RTX PRO 6000 Blackwell Server Edition GPUs. AWS positions these instances as a cost-effective option for generative AI inference and a high-performance platform for demanding graphics applications.

The G7e series offers significant upgrades over the previous G6e generation. Each Blackwell GPU provides 96 GB of memory, doubling the capacity of its predecessor, and a substantial increase in memory bandwidth. This allows a single GPU to handle AI models with up to 70 billion parameters. For even larger models, multiple GPUs within an instance can be linked using NVIDIA's GPUDirect P2P technology, which cuts communication latency between processors. A top-tier g7e.48xlarge instance packs eight GPUs, delivering a combined 768 GB of GPU memory.

Beyond raw graphics power, the instances feature major networking improvements. They offer four times the network bandwidth of G6e instances, making them viable for small multi-node clusters. Integration with AWS's Elastic Fabric Adapter and high-throughput storage options like Amazon FSx for Lustre is designed to speed up data loading and inter-node communication for distributed workloads.

G7e instances are currently offered in the US East (N. Virginia) and US East (Ohio) regions. They can be purchased on-demand, via Savings Plans, or as Spot Instances. Developers can begin using them through the AWS Management Console, CLI, or SDKs, and can employ AWS Deep Learning AMIs for machine learning projects. Support for Amazon SageMaker is expected soon.

Source: AWS