All news with #amazon sagemaker ai tag

🔒 Amazon SageMaker Unified Studio now supports custom IAM permissions boundaries so organizations enforcing Service Control Policies (SCPs) can provision projects without changing their security posture. When creating a project, SageMaker provisions three IAM roles — a project user role, an Amazon Bedrock service role, and a Bedrock Lambda execution role — and administrators can specify a permissions boundary in the Tooling blueprint configuration. The boundary is attached to all three roles at creation, satisfying SCP requirements and limiting role capabilities while allowing automatic project provisioning across all supported AWS Regions.

🔧 Amazon SageMaker HyperPod now offers troubleshooting skills that deliver expert AI/ML cluster diagnostics into coding assistants like Claude Code, Cursor, and Kiro. These skills guide natural-language diagnostics for GPU hardware faults, NCCL communication issues, and performance bottlenecks across distributed clusters. They automate evidence collection via AWS Systems Manager and provide actionable recommendations without requiring infrastructure changes.

🔧 Amazon SageMaker HyperPod now supports EFA-only network interfaces for cluster instance groups, allowing dedicated Elastic Fabric Adapter devices without attaching Elastic Network Adapters for IP networking. This reduces IP address consumption in VPC subnets and enables larger-scale distributed training clusters. To enable it, set efa-only in the ClusterNetworkInterface when creating or updating a HyperPod cluster via the API.

🚀 Amazon announces general availability of EC2 P5.48xl instances in Asia Pacific (Tokyo) for use with SageMaker notebook instances. These instances use NVIDIA H100 Tensor Core GPUs to accelerate deep learning and HPC workloads, promising up to 4x faster time to solution and up to 40% lower training costs versus previous GPU generations. Customers can leverage P5 instances to train and deploy complex LLMs and diffusion models for generative AI tasks. Developer guides provide setup instructions for JupyterLab and CodeEditor on SageMaker Studio and notebook instances.

🚀 Amazon EC2 P4de instances are now generally available on SageMaker notebook instances in Asia Pacific (Tokyo). These instances feature 8 NVIDIA A100 GPUs with 80GB HBM2e each, delivering 640GB total GPU memory and up to 60% better ML training performance versus P4d. Customers can expect reduced model training times, improved handling of high-resolution datasets, and about 20% lower training cost. See developer guides for setup details on SageMaker Studio and notebook instances.

🚀 Amazon announced the general availability of EC2 P6-B200 instances on SageMaker notebook instances in AWS US East (N. Virginia). These instances feature 8 NVIDIA Blackwell GPUs with 1440 GB GPU memory and 5th Gen Intel Xeon processors, offering up to 2x performance versus P5en for AI training. Customers can use them to develop and fine-tune large foundation models interactively in JupyterLab or CodeEditor on SageMaker Studio.

🚀 Amazon SageMaker notebook instances now support EC2 P5.4xl instances powered by NVIDIA H100 GPUs. These instances boost deep learning and HPC workloads, offering up to 4x faster time-to-solution and up to 40% lower training cost versus prior GPU generations. P5.4xl is available in multiple AWS regions including US East, US West, Asia Pacific, and South America; see AWS developer guides for setup instructions.

🚀 Amazon Web Services announces general availability of Amazon EC2 P5en.48xl instances on SageMaker notebook instances. These P5en instances feature 8 H200 GPUs with increased GPU memory and bandwidth versus H100, paired with custom 4th Gen Intel Xeon processors and Gen5 PCIe for higher CPU–GPU bandwidth. They also include third-generation EFA via Nitro v5, offering up to 3200 Gbps and latency improvements over prior P5 instances. P5en.48xl is currently available in US East (N. Virginia, Ohio), US West (Oregon), and Asia Pacific (Tokyo).

🚀 Amazon announces GA of EC2 P5en.48xl instances for SageMaker notebook instances, delivering advanced H200 GPUs paired with 4th Gen Intel Xeon processors. These instances provide increased GPU memory and bandwidth compared to P5, Gen5 PCIe between CPU and GPU, and faster EFA/Nitro networking to boost distributed training and inference. P5en.48xl is available in US East (N. Virginia, Ohio), US West (Oregon), and Asia Pacific (Tokyo) regions. Refer to the developer guides for setup and SageMaker Studio integration.

🚀 Amazon SageMaker notebook instances now support EC2 P5.4xl instances powered by NVIDIA H100 Tensor Core GPUs. These instances deliver up to 4x higher performance and up to 40% lower training cost versus prior-generation GPU instances, accelerating development of deep learning and generative AI models. P5.4xl is generally available across multiple AWS regions including US East, US West, Asia Pacific, and South America. Refer to developer guides for setup instructions in JupyterLab and CodeEditor on SageMaker Studio and notebook instances.

🔔 Amazon SageMaker HyperPod now supports minimum capacity requirements (MinCount) for Slurm clusters using continuous provisioning, ensuring that instance groups only enter InService when a specified minimum number of instances are provisioned. This is valuable for distributed training frameworks such as PyTorch FSDP, Megatron-LM, and NVIDIA NeMo that need a fixed node count, and for teams requiring baseline GPU counts to meet SLAs or cost targets. Specify MinInstanceCount in the CreateCluster or UpdateCluster API; groups remain in Creating or Updating until the threshold is met or a 3-hour rollback triggers. MinCount is available in all Regions supporting SageMaker HyperPod.

🧭 Amazon SageMaker Unified Studio IAM domains now include an interactive interface for creating and managing feature groups in SageMaker Feature Store, removing the need to write code for common feature management tasks. This launch makes feature management accessible to data scientists, ML engineers, and business analysts from a single collaborative environment. Users can discover and search existing features, create and modify feature groups, view definitions and schemas, and monitor data ingestion status without API calls. Features created elsewhere appear immediately in SageMaker Unified Studio when sharing the same IAM role, enabling seamless workflows across the ML lifecycle.

🔒 Amazon SageMaker Unified Studio now supports domain management for both Identity Center and IAM-based domains outside the AWS Console. Administrators and data management teams can create and manage projects, configure workforce identity, administer users and permissions, and set networking properties. VPC configuration and account associations are consistent across domain types and available in all Regions where Unified Studio is offered.

🛠️ Amazon SageMaker Unified Studio now adds business context, metadata, and data governance features for IAM-based domains. Customers can annotate AWS Glue Data Catalog tables with business names, descriptions, and README documentation, and use AI-generated metadata to automate cataloging. Teams can build business glossaries, define metadata form templates, and capture structured attributes like classification, retention, and ownership. These capabilities enable search, filtering by glossary or metadata fields, and access requests with automated Lake Formation permission grants, and are available in all regions where SageMaker Unified Studio is supported.

🔧 Amazon SageMaker Unified Studio now automatically creates Glue connections across subnets to enable job retries when a primary subnet becomes unavailable. Administrators define a domain VPC with multiple private subnets and the system provisions connectors for new projects so retries can run on alternate subnets without manual intervention. This reduces unplanned data-pipeline downtime and helps meet SLAs across AWS Regions where SageMaker Unified Studio is available.

🧩 Amazon SageMaker Inference now supports OpenAI-compatible APIs, enabling existing tools and frameworks like the OpenAI SDK, LangChain, and Strands Agents to connect directly to SageMaker endpoints. Switching requires only changing an endpoint URL, with no custom integration code or SDK wrappers. You can continue using your current authentication approach while choosing GPU instances, keeping data in your VPC, running open source or fine-tuned models, and leveraging auto-scaling policies. This capability is available today across multiple AWS regions with AWS credentials and automatic token refresh for production use.

🛠️ Amazon SageMaker Unified Studio now integrates data quality rule authoring and evaluation powered by AWS Glue Data Quality. Data engineers, analysts, and data scientists can define rules, run evaluations, and view results for both data at rest and data in transit. The feature supports catalog table checks and Visual ETL job evaluations to detect issues before they impact analytics or ML workloads.

🧾 Amazon SageMaker HyperPod now supports data capture for inference workloads, allowing organizations to record request and response payloads for monitoring, compliance, debugging, and offline analysis. You can capture traffic at the SageMaker endpoint, load balancer, or model pod and combine layers for richer observability. Captured data is delivered asynchronously to Amazon S3 with configurable sampling and encryption using customer-managed AWS KMS keys and is designed to never block inference. Enable data capture via the HyperPod Inference Operator or SageMaker JumpStart.

🚀 Amazon SageMaker Studio IDEs, including JupyterLab and Code Editor, now support GPU capacity reservations via SageMaker Flexible Training Plans (FTP), offering predictable access to high-performance resources and up to 65% cost savings versus On‑Demand. FTP provides a self-serve procurement flow to select instance type, reservation length, and start date. Studio apps can be launched using the purchased plan from the Instance dropdown, with automatic provisioning and proactive expiration notifications to protect work.

🚀 Amazon SageMaker AI now supports serverless customization for the Qwen3.6 27B parameter model using supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT). This extends SageMaker's existing fine-tuning support for Qwen3.5 and other open-weight models. Serverless customization removes infrastructure management—SageMaker handles provisioning and orchestration—so teams pay only for what they use. The feature is available in US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and EU (Ireland).