aztk/docs/60-gpu.md

# GPU

Use GPUs to accelerate your Spark applications. When using a [GPU enabled Azure VM](https://azure.microsoft.com/en-us/pricing/details/batch/), your docker image will contain CUDA-8.0 and cuDnn-6.0 by default. See [Docker Image](./12-docker-image.html) for more information about the AZTK Docker images.

[NOTE: Azure does not have GPU enabled VMs in all regions. Please use this [link](https://azure.microsoft.com/en-us/pricing/details/batch/) to make sure that your Batch account is in a region that has GPU enabled VMs]

AZTK uses Nvidia-Docker to expose the VM's GPU(s) inside the container. Nvidia drivers (ver. 384) are installed at runtime.


### Tutorial

Create a cluster specifying a GPU enabled VM
```sh
aztk spark cluster create --id gpu-cluster --vm-size standard_nc6 --size 1
```

Submit your an application to the cluster that will take advantage of the GPU
```sh
aztk spark cluster submit --id gpu-cluster --name gpu-app ./examples/src/main/python/gpu/nubma_example.py
```
### Installation Location
By default, CUDA is installed at `/usr/local/cuda-8.0`.
Feature: Spark GPU (#206) * conditionally install and use nvidia-docker * status statements, and -y flag for install * add example, remove unnecessary ppa * rename custom script, remove print statement, update example * add Dockerfile * fix path in Dockerfile * update Docker images to use service account * updated docs, changed default docker repo for gpu skus * make timing statements more verbose * remove unnecessary script * added gpu docs * fix up docs and numba example 2017-12-05 00:28:05 +03:00			`# GPU`

Feature: Readthedocs support (#497) 2018-04-27 00:03:45 +03:00			`Use GPUs to accelerate your Spark applications. When using a [GPU enabled Azure VM](https://azure.microsoft.com/en-us/pricing/details/batch/), your docker image will contain CUDA-8.0 and cuDnn-6.0 by default. See [Docker Image](./12-docker-image.html) for more information about the AZTK Docker images.`
Feature: Spark GPU (#206) * conditionally install and use nvidia-docker * status statements, and -y flag for install * add example, remove unnecessary ppa * rename custom script, remove print statement, update example * add Dockerfile * fix path in Dockerfile * update Docker images to use service account * updated docs, changed default docker repo for gpu skus * make timing statements more verbose * remove unnecessary script * added gpu docs * fix up docs and numba example 2017-12-05 00:28:05 +03:00
Update 60-gpu.md (#253) * Update 60-gpu.md make sure is available in region * Update 60-gpu.md 2017-12-08 01:11:53 +03:00			`[NOTE: Azure does not have GPU enabled VMs in all regions. Please use this [link](https://azure.microsoft.com/en-us/pricing/details/batch/) to make sure that your Batch account is in a region that has GPU enabled VMs]`

Feature: Spark GPU (#206) * conditionally install and use nvidia-docker * status statements, and -y flag for install * add example, remove unnecessary ppa * rename custom script, remove print statement, update example * add Dockerfile * fix path in Dockerfile * update Docker images to use service account * updated docs, changed default docker repo for gpu skus * make timing statements more verbose * remove unnecessary script * added gpu docs * fix up docs and numba example 2017-12-05 00:28:05 +03:00			`AZTK uses Nvidia-Docker to expose the VM's GPU(s) inside the container. Nvidia drivers (ver. 384) are installed at runtime.`


			`### Tutorial`

			`Create a cluster specifying a GPU enabled VM`
			```sh
			`aztk spark cluster create --id gpu-cluster --vm-size standard_nc6 --size 1`
			```

			`Submit your an application to the cluster that will take advantage of the GPU`
			```sh
			`aztk spark cluster submit --id gpu-cluster --name gpu-app ./examples/src/main/python/gpu/nubma_example.py`
			```
			`### Installation Location`
			By default, CUDA is installed at `/usr/local/cuda-8.0`.