GPUs with Dataflow

Dataflow GPUs bring the accelerated benefits directly to your stream or batch data processing pipeline. Use Dataflow to simplify the process of getting data to the GPU and to take advantage of data locality. At the same time, get all the benefits of the fully managed Dataflow system: host provisioning, autoscaling, fault tolerance, and more.
Follow this tutorial to see how to build a custom container image and run a Dataflow pipeline with GPUs.
Follow this tutorial to build a PyTorch pipeline and run it on Dataflow with GPUs.
The L4 GPU type is useful for running machine learning inference pipelines.
Follow this tutorial to build a TensorFlow pipeline and run it on Dataflow with GPUs.

Resources

See an example of a developer workflow for building pipelines using GPUs.
Jobs using GPUs incur charges as specified in the Dataflow pricing page.
If you run into problems running your Dataflow job with GPUs, follow these troubleshooting steps.