Performance and cost optimization best practices for machine learning
Donna Schut
Head of Technical Solutions Management, Applied AI Engineering
Khalid Salama
Staff Machine Learning Solutions Architect
Cloud computing provides the power and speed needed for Machine Learning (ML), and allows you to easily scale up and down. However, this also means that costs may spin out of control if you don't plan ahead, which is especially fraught now, given that businesses are particularly cost conscious.
To use Google Cloud effectively for ML, then, it’s important to follow best practices to optimize for performance and costs. To help you do just that, we published a new set of best practices—based on our experience working with advanced ML customers—on how you can enhance the performance and decrease the costs of your ML workloads on Google Cloud, from experimentation to production. The guide covers various Smart Analytics and Cloud AI services in different phases of the ML process, as illustrated in the diagram below, namely:
Experimentation with AI Platform Notebooks
Training with AI Platform Training
Serving with AI Platform Prediction
- Orchestration with AI Platform Pipelines
We also provide best practices for monitoring performance and managing the cost of ML projects with Google Cloud tools. Are you ready to optimize your ML workloads? Check out the Machine Learning Performance and Cost Optimization Best Practices to get started.
Acknowledgements: We’d like to thank Andrew Stein (Product Manager, Cloud Accelerators), Chad Jennings (Product Manager, BigQuery), Henry Tappen (Product Manager, Cloud AI), Karthik Ramachandran (Product Manager, Cloud AI), Lak Lakshmanan (Head of Data Analytics and AI Solutions), Mark Mirchandani (Developer Advocacy, Cost Management), Kannappan Sirchabesan (Strategic Cloud Engineer, Data and Analytics), Mehran Nazir (Product Manager, Dataflow), and Shan Kulandaivel (Product Manager, Dataflow) for their contributions to the best practice documentation.
