AWS

AWS IAM Identity Center now allows you to disable AWS account access management during initial organization instance setup. This enables using the service solely for workforce identity integration with AWS applications, decoupling app SSO from account-level permissions. The change applies only to new instances and does not retroactively affect existing configurations.

• New org instances can now decouple app SSO from AWS account access management
• Use Identity Center for application access only, without managing account permissions
• Feature is available only at initial instance configuration stage
• Existing Identity Center instances remain unaffected by this change

Amazon Timestream for InfluxDB now supports on-demand and scheduled backups for both InfluxDB 2 and 3 engines. Users can manage up to four automated backup configurations with custom retention periods or trigger one-time snapshots via the console, CLI, or API. Restores can be applied to new resources or overwrite existing ones, giving teams direct control over data protection strategies.

• Supports both InfluxDB 2 and 3 engines across console, CLI, and API.
• Allow up to four automated backup schedules with flexible retention.
• Enable on-demand backups before risky migrations or config changes.
• Restore to a new resource or overwrite existing data on demand.

Amazon EC2 R8i and R8i-flex instances are now available in the Europe (Milan) region, powered by custom Intel Xeon 6 processors exclusive to AWS. These new instances provide up to 20% higher performance than the previous R7i generation, with significant gains in memory bandwidth and price-performance. Specific workloads see even larger improvements, including 30% faster PostgreSQL performance and 60% faster NGINX web applications.

• R8i instances use custom Intel Xeon 6 processors for highest cloud memory bandwidth.
• PostgreSQL workloads see up to 30% performance improvement over R7i instances.
• NGINX web apps are up to 60% faster, boosting web serving efficiency.
• R8i-flex offers memory-optimized options for specific high-memory workloads.
• Migration to Milan region now supports these latest Intel-based compute options.

AWS has added authorizer log delivery for Amazon MSK Provisioned clusters, including Standard and Express brokers, at no extra cost. This feature captures details of denied access requests, such as client IP addresses and targeted APIs, to aid in troubleshooting and security compliance. Logs can be routed to CloudWatch Logs, S3, or Data Firehose via the console or CLI.

• Free authorizer logs for Provisioned MSK clusters (Standard/Express)
• Capture denied request details including client IP and API
• Route logs to CloudWatch, S3, or Data Firehose
• Enable via console or CLI for new and existing clusters

Amazon ECS now allows fractional GPU allocation using EC2 G6f instances, enabling workloads to use GPU partitions as small as 1/8 of an NVIDIA L4 Tensor Core. This feature targets small-model AI inference, experimentation, and graphics rendering by allowing precise right-sizing of container resources. Users can specify fractional requirements like 0.125, 0.25, or 0.5 GPUs directly in the task definition container settings.

• Right-size containers for small AI inference and experiments without provisioning full GPUs.
• Set fractional GPU requests (0.125, 0.25, 0.5) directly in ECS task definitions.
• Reduce infrastructure costs by avoiding over-provisioning of GPU memory and compute.
• Works with NVIDIA L4 Tensor Core GPUs on G6f instances for efficient resource utilization.

Amazon DynamoDB now offers general availability for native vector search, allowing engineers to index and query embeddings directly within DynamoDB tables. The feature delivers single-digit millisecond latency with over 99% recall, designed to handle datasets scaling into the trillions of vectors. Users can store embeddings alongside standard attributes and utilize approximate nearest neighbor searches using models from sources like Amazon Bedrock.

• Native vector search eliminates the need for external vector databases for many AI workloads.
• Single-digit millisecond latency and 99%+ recall scale to trillions of vectors.
• Embeddings can be stored alongside standard DynamoDB attributes in the same table.
• Supports approximate nearest neighbor (ANN) search via vector indexes.
• Integrates with Amazon Bedrock models for generating embeddings.