Note: The following considerations apply to this GA offering:

• Jobs that use GPUs incur charges as specified in the Dataflowpricing page.
• To use GPUs, your Dataflow job must useDataflow Runner v2.

This page provides background information on how GPUs work withDataflow, including information about prerequisites and supportedGPU types.

Using GPUs in Dataflow jobs lets you acceleratesome data processing tasks. GPUs can perform certain computations fasterthan CPUs. These computations are usually numeric or linear algebra,often used in image processing and machine learning use cases. Theextent of performance improvement varies by the use case, type of computation,and amount of data processed.

Prerequisites for using GPUs in Dataflow

• To use GPUs with your Dataflow job, you must use Runner v2.
• Dataflow runs user code in worker VMs inside a Docker container. These worker VMs runContainer-Optimized OS. For Dataflow jobs to use GPUs, you need the following prerequisites:
  • GPU drivers are installed on worker VMs and accessible to the Docker container. For more information, seeInstall GPU drivers.
  • GPU libraries required by your pipeline, such asNVIDIA CUDA-X libraries or theNVIDIA CUDA Toolkit, are installed in the custom container image. For more information, seeConfigure your container image.
• Because GPU containers are typically large, to avoidrunning out of disk space, increase the defaultboot disk size to 50 gigabytes or more.

Pricing

Jobs using GPUs incur charges as specified in the Dataflowpricing page.

Availability

Note: TPUs are also supported with Dataflow. For more information,seeDataflow support for TPUs.

The following GPU types are supported with Dataflow:

GPU type	worker_accelerator string
NVIDIA® L4	nvidia-l4
NVIDIA® A100 40 GB	nvidia-tesla-a100
NVIDIA® A100 80 GB	nvidia-a100-80gb
NVIDIA® Tesla® T4	nvidia-tesla-t4
NVIDIA® Tesla® P4	nvidia-tesla-p4
NVIDIA® Tesla® V100	nvidia-tesla-v100
NVIDIA® Tesla® P100	nvidia-tesla-p100
NVIDIA® H100	nvidia-h100-80gb
NVIDIA® H100 Mega	nvidia-h100-mega-80gb

For more information about each GPU type, including performance data, seeCompute Engine GPU platforms.

For information about available regions and zones for GPUs, seeGPU regions and zones availabilityin the Compute Engine documentation.

Recommended workloads

The following table provides recommendations for which type of GPU to use fordifferent workloads. The examples in the table are suggestions only, and youneed to test in your own environment to determine the appropriate GPU type foryour workload.

For more detailed information about GPU memory size, feature availability, andideal workload types for different GPU models, see theGeneral comparison charton the GPU platforms page.

Workload	A100, H100	L4	T4
Model fine tuning	Recommended
Large model inference	Recommended	Recommended
Medium model inference		Recommended	Recommended
Small model inference		Recommended	Recommended

What's next

• See an example of adeveloper workflow for building pipelines that use GPUs.
• Learn how torun an Apache Beam pipeline on Dataflow with GPUs.
• Work throughProcessing Landsat satellite images with GPUs.