AWS Cuts Prices for SageMaker GPU Instances
Amazon Web Services is cutting prices on its SageMaker managed
service for machine learning and deep learning as it attracts more financial services, healthcare and retail customers building and training ML models in production.
The cloud giant (NASDAQ: AMZN) said Wednesday (Oct. 7) it is reducing prices for GPU instances running SageMaker by as much as 18 percent. Reminiscent of earlier price cuts as AWS battled Microsoft Azure and Google (NASDAQ: GOOGL) for public cloud dominance, the reductions for SageMaker reflect the growing number of enterprise options for building, training and deploying machine and deep learning as production workloads.
Released in late 2017, SageMaker was among the first model trainers out of the gate. Since then AWS has expanded the ecosystem to include tools for building and managing training data sets along with an integrated development environment dubbed SageMaker Studio. The IDE allows developers to collect and store code, notebooks, data sets, settings and project folders in a single place.
In June, AWS updated its SageMaker data labeling service called Ground Truth introduced in 2018 with a workflow for labeling point clouds, a set of data points generated by tools like 3D scanners or lidar sensors. Among the applications is labeling huge 3D data sets used to train models incorporated into self-driving car navigation systems. Those data sets can grow to hundreds of megabytes, making labeling extremely arduous.
The price reductions announced this week cover cloud GPUs running as Amazon Elastic Cloud Compute “P2” and “P3” instances. General purpose P2 instances provide up to 16 Nvidia K80 GPUs, 64 virtual CPUs and 732 Gb of host memory. Higher-end P3 instances deliver a maximum of 8 Nvidia V100 Tensor Core GPUs and up to 100 Gbps of networking throughput.
AWS said price reductions for those SageMaker machine learning instances range from 11 percent to 18 percent. The price cuts took effect on Oct. 1 for all SageMaker components and cover four North American regions, three EU regions and five in the Asia-Pacific. Also included are the AWS GovCloud.
Along with cheaper GPU instances, AWS continues to promote SageMaker and its expanding set of tools such as machine learning libraries as a way to reduce costs as training jobs grow in complexity. A debugging feature, for example, is designed to spot problems in machine learning training jobs. Faulty workloads are terminated early, and model information gathered during training can be used to determine the root cause.