New feature
GA
Databricks Runtime for Machine Learning
Databricks Runtime for Machine Learning is a Databricks compute / bi capability, introduced May 2018.
A variant of Databricks Runtime for classic compute that comes pre-built with machine learning and deep learning infrastructure and the most common ML and DL libraries (with a Databricks-tested set of top-tier libraries updated every release), enabled by ticking the Machine learning checkbox when you create compute.
- Its May 2018 debut was a Beta pitched on one party trick: a cluster where TensorFlow, CUDA, and cuDNN already agreed with each other, plus distributed TensorFlow training via Horovod - and it was not available on Community Edition accounts.
- It spent almost a year in Beta across versions 4.1 to 5.2 before 5.3 ML, in April 2019, earned the release note line "we have achieved our first GA of Databricks Runtime ML!"
- From release 6.1 in October 2019 the plain Databricks Runtime went CPU-only, so the ML runtime became the way to get a GPU cluster - until the docs started recommending serverless AI Runtime over it for custom deep learning on GPUs.
- Category
- Compute / BI
- Also known as
- Databricks Runtime ML, DBR ML, Databricks Runtime for ML, ML runtime, Machine learning checkbox
- Verified
- 2026-09-27
Sources
- Official Official Databricks / Microsoft docs
- Official Databricks release notes: Databricks Runtime 4.1 for Machine Learning (Beta), May 18, 2018
- Official Databricks release notes: Databricks Runtime 5.3 ML, the first GA (April 3, 2019)
- Official Databricks docs: for custom deep learning on GPUs, Databricks recommends serverless AI Runtime instead (a separate offering, not a rename)