AWS sagemaker documentation change
Summary
Added documentation for UltraServers in SageMaker AI including instance types, regional availability, and operational considerations
Security assessment
The changes primarily introduce new UltraServer instance types and operational guidance. While they mention fault resiliency and recovery mechanisms, these are reliability features rather than security fixes. No specific vulnerabilities or security controls are addressed.
Diff
diff --git a/sagemaker/latest/dg/reserve-capacity-with-training-plans.md b/sagemaker/latest/dg/reserve-capacity-with-training-plans.md index 11e698f9d..2faab0e59 100644 --- a//sagemaker/latest/dg/reserve-capacity-with-training-plans.md +++ b//sagemaker/latest/dg/reserve-capacity-with-training-plans.md @@ -5 +5 @@ -What is SageMaker training plansBenefitsReservationUser workflowSupported instance types, AWS Regions, and pricingSearch behaviorConsiderations +What are SageMaker training plansBenefitsReservationUser workflowSupported instance types, AWS Regions, and pricingUltraServers in SageMaker AISearch behaviorConsiderations @@ -9 +9 @@ What is SageMaker training plansBenefitsReservationUser workflowSupported instan -Amazon SageMaker training plans is a capability that allows you to reserve and help maximize the use of GPU capacity for large-scale AI model training workloads. This feature provides access to highly sought-after instance types that cover a range of GPU-accelerated computing options, including the latest NVIDIA GPU technologies and AWS Trainium chips. With SageMaker training plans, you can secure predictable access to these high-demand, high-performance computational resources within your specified timelines and budgets, without the need to manage underlying infrastructure. This flexibility is particularly valuable for organizations dealing with the challenges of acquiring and scheduling these oversubscribed compute instances for their mission-critical AI workloads. +Amazon SageMaker training plans is a capability that allows you to reserve and help maximize the use of GPU capacity for large-scale AI model training workloads. This feature provides access to highly sought-after instance types that cover a range of GPU-accelerated computing options, including the latest NVIDIA GPU technologies and AWS trainium chips. With SageMaker training plans, you can secure predictable access to these high-demand, high-performance computational resources within your specified timelines and budgets, without the need to manage underlying infrastructure. This flexibility is particularly valuable for organizations dealing with the challenges of acquiring and scheduling these oversubscribed compute instances for their mission-critical AI workloads. @@ -11 +11 @@ Amazon SageMaker training plans is a capability that allows you to reserve and h -## What is SageMaker training plans +## What are SageMaker training plans @@ -137,0 +138,11 @@ Training plans support reservations for the following specific high-performance + * **ml.c6i-32xlargesc** + + + + +**UltraServers** + + * **ml.p6e-gb200.36xlarge** + + * **ml.p6e-gb200.72xlarge** + @@ -159,0 +171,29 @@ The availability across multiple regions allows to choose the most suitable loca +## UltraServers in SageMaker AI + +UltraServers in SageMaker AI offer a set of instances interconnected via a high bandwidth network domain. For example, the P6e-GB200 UltraServer connects up to 18 `p6e-gb200.36xlarge` instances under one NVIDIA NVLink domain. With 4 NVIDIA Blackwell GPUs per instance, each P6e-GB200 UltraServer supports 72 GPUs, so you can run your largest AI workloads with high performance on SageMaker AI. + +When you use UltraServers with SageMaker AI, you get performance combined with SageMaker AI's managed infrastructure, built-in fault resiliency features, integrated monitoring capabilities, and native integration with other SageMaker AI and AWS services. This integration allows you to focus on model development and deployment while SageMaker AI handles the undifferentiated heavy lifting of managing AI infrastructure. + +###### Note + +UltraServers are available only in the Dallas Local Zone (us-east-1-dfw-2a), which is an extension of the US East (N. Virginia) Region. For more information, see [ Getting started with AWS Local Zones](https://docs.aws.amazon.com/local-zones/latest/ug/getting-started.html) + +### Considerations + +Consider the following when using UltraServers with SageMaker AI: + + * You can use UltraServers for both [ SageMaker HyperPod](https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-hyperpod.html) and [ SageMaker training jobs](https://docs.aws.amazon.com/sagemaker/latest/dg/train-model.html). + + * You can only purchase UltraServers in full units. For more information about instance and pricing information, see Amazon SageMaker HyperPod flexible training plans in [Amazon SageMaker AI pricing](https://aws.amazon.com/sagemaker-ai/pricing/). + + * If you're using UltraServers with HyperPod, HyperPod automatically adds topology labels to your resources to help you with resource allocation. For more information, see [ Using topology-aware scheduling in Amazon SageMaker HyperPod](https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-hyperpod-topology.html). + + * SageMaker AI and UltraServers offer various capabilities that enhance the resiliency of your workloads, including preemptive checks and automatic fault detection and mitigation. Depending on what the issue is, SageMaker AI can run actions to recover your workloads, such as restarting instances, replacing failed instances with spares, and replacing failed UltraServers. + + * For added resilience, you can configure instances within an UltraServer to be used as spares. Keeping a spare instance within the UltraServer ensures that SageMaker AI can quickly respond to an instance failure while minimizing any impact to your jobs. We recommend that you keep one spare instance per UltraServer. You don't have to reserve any spare instances, but this might hinder support options and slow down failure recovery. You purchase UltraServers by wholes, so the number of spares that you reserve doesn't affect pricing. + + * To see the status and instances within an UltraServer, use the [ ListTrainingPlans](https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_ListTrainingPlans.html) API operation or the AWS console to see training plans. Using these tools, you can see the total number of available instances, instances currently in use, unhealthy instances, the number of configured spares, and other information. Possible health statuses are `ok`, `impaired`, and `insufficient-data`. + + + +