The DLVM is a specially configured variant of the Data Science VM Dsvm that is custom made to help users jump start Deep Learning on Azure GPU VMs. The DLVM uses the same underlying VM images of the DSVM and hence comes with the same set of data science tools and deep learning frameworks as the base VM.

DLVM VMs specifically a for GPU based deep learning workloads and improve the product discoverability as a premier deep learning VM environment on Azure.

Why I think this is perfect for education

• Ease of Provisioning: A specially built ARM template that helps user directly choose the GPU class VMs as we only allow provisioning of DLVM on NC Series machines. The template also allows user to choose between the Windows 2016 DSVM image and Ubuntu DSVM image during the provisioning of the VM.

So the scripting of multiple machine deployment is pretty straight forward we have some bulk deployment scripts available at https://github.com/MSFTImagine/computerscience/tree/master/Scripts/Bulk-Sandbox-Deployment-Automation-Bash

• New Tools and Frameworks: On the underlying DSVM images, we now include Chainer 3.0 RC (both Windows, Ubuntu), Visual Object Tagging Tool (VOTT) on Windows, Keras (Windows)  with TF, MS Cognitive Toolkit, Theano backend), Dlib, gensim, LightGBM.
  
  
• End to End Deep Learning samples: We have now augmented the individual framework samples with end to end samples  comparing and migrating between deep learning frameworks, object detection solution how to guide, Text analytics to classify Amazon reviews, Text analytics with Biomedical entity extraction. In the DLVM ARM template, the samples are automatically downloaded (git cloned) when the VM is created.
  
  
• Azure Machine Learning Support: The DSVM/DLVM can be used as a dev/experimentation for Azure ML.  It can be used to locally train models on GPUs.  The Ubuntu DSVM can be used as a remote docker execution target.
  
  
• Batch Service Integration: DSVM & DLVM image VM images available for deployment under Azure Batch https://azure.microsoft.com/en-us/services/batch/.

Some of the other non deep learning highlights on the latest release of the underlying DSVM images are:

• Spark standalone on Windows 2016 DSVM and local Spark Python3 notebook kernel. Now you can develop and test PySpark code locally on Windows DSVM (standalone or within AzureML workbench).
• MMLSpark library on both Windows 2016 and Ubuntu DSVM and ability to run MMLSpark on local Spark standalone instance.
• DeepWater and Sparkling Water (on Ubuntu only).
• Azure ML Workbench bootstrap installer on Windows 2016 DSVM (Clicking on a desktop icon installs the whole Azure ML Workbench and CLI locally)
• PyCharm is now on Windows 2016 DSVM
• Geo Spatial DSVM This extensions adds on ESRI’s ArcGIS Pro desktop software on top of Windows 2016 DSVM along with the Python and R bridges.

Getting started

DSVM Product Web page: http://aka.ms/dsvm

Provisioning the DLVM: http://aka.ms/dlvm

Documentation: http://aka.ms/dlvm/doc

Deep Learning and AI frameworks http://aka.ms/dsvm/discover 

Tooling and IDE

Additionally we have released the Visual Studio Code AI extension. VS Code Tools for AI is a cross-platform extension that supports deep learning frameworks including Microsoft Cognitive Toolkit (CNTK), Google TensorFlow and more https://blogs.msdn.microsoft.com/uk_faculty_connection/2017/09/28/visual-studio-code-tools-for-ai-extension/