GPU-enabled compute

Overview

Azure Databricks supports compute accelerated with graphics processing units (GPUs). This article describes how to create compute with GPU-enabled instances and describes the GPU drivers and libraries installed on those instances.

To learn more about deep learning on GPU-enabled compute, see Deep learning.

Create a GPU compute

Creating a GPU compute is similar to creating any compute. You should keep in mind the following:

• The Databricks Runtime Version must be a GPU-enabled version, such as Runtime 13.3 LTS ML (GPU, Scala 2.12.15, Spark 3.4.1).
• The Worker Type and Driver Type must be GPU instance types.

Supported instance types

Azure Databricks supports the following instance types:

• NC v3 instance type series: Standard_NC6s_v3, Standard_NC12s_v3, Standard_NC24s_v3
• NC T4 v3 instance type series: Standard_NC4as_T4_v3, Standard_NC8as_T4_v3, Standard_NC16as_T4_v3, Standard_NC64as_T4_v3

• NV A10 v5 instance type series: Standard_NV36ads_A10_v5, Standard_NV36adms_A10_v5, Standard_NV72ads_A10_v5

See Azure Databricks Pricing for an up-to-date list of supported GPU instance types and their availability regions. Your Azure Databricks deployment must reside in a supported region to launch GPU-enabled compute.

GPU scheduling

GPU scheduling distributes Spark tasks efficiently across a large number of GPUs.

Databricks Runtime supports GPU-aware scheduling from Apache Spark 3.0. Azure Databricks preconfigures it on GPU compute.

User-defined GPU scheduling is only available for Databricks Runtime 7.1 and above. For previous versions of Databricks Runtime, Databricks automatically configures GPU compute so that there is at most one running task per node. That way, the task can use all GPUs on the node without running into conflicts with other tasks.

GPU scheduling for AI and ML

spark.task.resource.gpu.amount is the only Spark config related to GPU-aware scheduling that you may need to configure. The default configuration uses one GPU per task, which is a good baseline for distributed inference workloads and distributed training if you use all GPU nodes.

To reduce communication overhead during distributed training, Databricks recommends setting spark.task.resource.gpu.amount to the number of GPUs per worker node in the compute Spark configuration. This creates only one Spark task for each Spark worker and assigns all GPUs in that worker node to the same task.

To increase parallelization for distributed deep learning inference, you can set spark.task.resource.gpu.amount to fractional values such as 1/2, 1/3, 1/4, … 1/N. This creates more Spark tasks than there are GPUs, allowing more concurrent tasks to handle inference requests in parallel. For example, if you set spark.task.resource.gpu.amount to 0.5, 0.33, or 0.25, then the available GPUs will be split among double, triple, or quadruple the number of tasks.

GPU indices

For PySpark tasks, Azure Databricks automatically remaps assigned GPU(s) to zero-based indices. For the default configuration that uses one GPU per task, you can use the default GPU without checking which GPU is assigned to the task. If you set multiple GPUs per task, for example, 4, the indices of the assigned GPUs are always 0, 1, 2, and 3. If you do need the physical indices of the assigned GPUs, you can get them from the CUDA_VISIBLE_DEVICES environment variable.

If you use Scala, you can get the indices of the GPUs assigned to the task from TaskContext.resources().get("gpu").

NVIDIA GPU driver, CUDA, and cuDNN

Azure Databricks installs the NVIDIA driver and libraries required to use GPUs on Spark driver and worker instances:

• CUDA Toolkit, installed under /usr/local/cuda.
• cuDNN: NVIDIA CUDA Deep Neural Network Library.
• NCCL: NVIDIA Collective Communications Library.

The version of the NVIDIA driver included is 535.54.03, which supports CUDA 11.0. For the NV A10 v5 instance type series, the version of the NVIDIA driver included is 535.154.05.

For the versions of the libraries included, see the release notes for the specific Databricks Runtime version you are using.

NVIDIA End User License Agreement (EULA)

When you select a GPU-enabled "Databricks Runtime Version" in Azure Databricks, you implicitly agree to the terms and conditions outlined in the NVIDIA EULA with respect to the CUDA, cuDNN, and Tesla libraries, and the NVIDIA End User License Agreement (with NCCL Supplement) for the NCCL library.

Databricks Container Services on GPU compute

You can use Databricks Container Services on compute with GPUs to create portable deep learning environments with customized libraries. See Customize containers with Databricks Container Service for instructions.

To create custom images for GPU compute, you must select a standard runtime version instead of Databricks Runtime ML for GPU. When you select Use your own Docker container, you can choose GPU compute with a standard runtime version. The custom images for GPU are based on the official CUDA containers, which is different from Databricks Runtime ML for GPU.

When you create custom images for GPU compute, you cannot change the NVIDIA driver version because it must match the driver version on the host machine.

The databricksruntime Docker Hub contains example base images with GPU capability. The Dockerfiles used to generate these images are located in the example containers GitHub repository, which also has details on what the example images provide and how to customize them.