AWS

Amazon ElastiCache now supports Graviton4-based M8g, R8g, and C8gn node families for both Valkey and Memcached engines. These new nodes deliver up to 47% higher throughput and 43% lower P99 latency compared to Graviton3 equivalents, with up to 31% better price-performance on on-demand pricing. The M8g family also provides up to 20% more memory per node size, while C8gn nodes offer 200 Gbps network bandwidth for network-intensive workloads.

• Graviton4 nodes boost throughput by 47% and cut P99 latency by 43% vs Graviton3.
• M8g nodes provide up to 20% more memory at the same instance size.
• C8gn nodes deliver 200 Gbps bandwidth for high-network throughput workloads.
• Supported engines include both Valkey and Memcached on ElastiCache.

Amazon DynamoDB has generally available vector search, enabling real-time indexing and querying of vector embeddings at scale. The service promises single-digit millisecond latency with 99%+ recall, addressing the traditional trade-offs between speed, scale, and accuracy for datasets reaching trillions of vectors. Developers can store embeddings alongside standard attributes and use approximate nearest neighbor searches with indexes of their choice.

• Native vector index supports approximate nearest neighbor searches in DynamoDB.
• Latency targets single-digit milliseconds with 99%+ recall at any scale.
• Embeddings can be generated via custom models or Amazon Bedrock.
• Vectors are stored alongside standard DynamoDB attributes in the same table.

Amazon ECS now allows task definitions to request fractional GPU resources, supporting partitions as small as one-eighth of an NVIDIA L4 Tensor Core GPU. This feature targets G6f instances and enables right-sizing for workloads like small-model inference or experimentation that do not require full GPU capacity. Users specify these allocations directly in container definitions using values such as 0.125, 0.25, or 0.5 to optimize cost and resource utilization.

• Right-size containers for AI inference or graphics without paying for full GPUs.
• Request 1/8, 1/4, or 1/2 GPU allocations in ECS task definitions.
• Reduces infrastructure costs for small-model experimentation workloads.
• Requires Amazon EC2 G6f instances with NVIDIA L4 Tensor Core GPUs.

AWS Lambda now offers scalable egress and ingress bandwidth for functions running outside a VPC, scaling linearly with memory from 625 Mbps at 2 GB to 3,000 Mbps at 10 GB. This eliminates the previous hard cap that constrained latency-sensitive workloads transferring large datasets. The update allows higher memory allocations to directly yield proportional network throughput improvements.

• Non-VPC Lambda functions now scale network bandwidth proportionally with memory allocation.
• Egress/ingress speeds range from 625 Mbps (2 GB) to 3 Gbps (10 GB).
• Removes previous 625 Mbps bottleneck for data-intensive, latency-sensitive workloads.
• No VPC configuration required to access increased throughput capabilities.

AWS has updated IAM Identity Center to allow optional management of AWS account access during the initial configuration of new organization instances. This enables organizations to use the service solely for workforce identity federation and application single sign-on, decoupling it from native AWS account permission management. The change applies only to newly created instances and does not alter existing deployments.

• New org instances can now skip AWS account access management, focusing only on application SSO.
• Decouple workforce identity federation from native AWS account permissioning for new setups.
• Existing IAM Identity Center instances remain unaffected by this configuration option.
• Streamlines access management for app owners and end users via consistent SSO experiences.

Amazon EC2 G7 instances powered by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs are now available in the Europe (Spain) Region. These instances offer up to 4.6x AI inference performance and 2.1x graphics performance compared to the previous G6 generation. The new hardware accelerates workloads including AI inference, real-time graphics rendering, and GPU-accelerated data analytics.

• G7 instances in EU (Spain) use NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs.
• AI inference performance improves up to 4.6x over G6 instances.
• Graphics performance increases up to 2.1x compared to G6.
• Suitable for LLM inference, video transcoding, and spatial computing.
• Enables local deployment of AI models without data residency constraints.