What is the Azure Data Science Virtual Machine for Linux and Windows?
The Data Science Virtual Machine (DSVM) is a customized VM image on the Azure cloud platform built specifically for doing data science. It has many popular data science tools preinstalled and preconfigured to jump-start building intelligent applications for advanced analytics.
The DSVM is available on:
- Windows Server 2019
- Windows Server 2022
- Ubuntu 20.04 LTS
Additionally, we're excited to offer Azure DSVM for PyTorch, which is an Ubuntu 20.04 image from Azure Marketplace that is optimized for large, distributed deep learning workloads. It comes preinstalled and validated with the latest PyTorch version to reduce setup costs and accelerate time to value. It comes packaged with various optimization functionalities (ONNX Runtime, DeepSpeed, MSCCL, ORTMoE, Fairscale, Nvidia Apex), and an up-to-date stack with the latest compatible versions of Ubuntu, Python, PyTorch, CUDA.
Comparison with Azure Machine Learning
The DSVM is a customized VM image for Data Science but Azure Machine Learning is an end-to-end platform that encompasses:
- Fully Managed Compute
- Compute Instances
- Compute Clusters for distributed ML tasks
- Inference Clusters for real-time scoring
- Datastores (for example Blob, ADLS Gen2, SQL DB)
- Experiment tracking
- Model management
- Environments (manage conda and R dependencies)
- Pipelines (automate End-to-End Data science workflows)
Comparison with Azure Machine Learning Compute Instances
Azure Machine Learning Compute Instances are a fully configured and managed VM image whereas the DSVM is an unmanaged VM.
Key differences between these:
|Azure Machine Learning
|Python, R, Julia, SQL, C#,
Java, Node.js, F#
|Python and R
|Pre-Configured GPU Option
|Scale up option
(requires additional configuration)
(requires additional configuration)
|Jupyter(lab), VS Code,
Visual Studio, PyCharm, Juno,
Power BI Desktop, SSMS,
Microsoft Office 365, Apache Drill
Sample use cases
Here's some common use cases for DSVM customers.
Short-term experimentation and evaluation
Deep learning with GPUs
In the DSVM, your training models can use deep learning algorithms on hardware that's based on graphics processing units (GPUs). By taking advantage of the VM scaling capabilities of the Azure platform, the DSVM helps you use GPU-based hardware in the cloud according to your needs. You can switch to a GPU-based VM when you're training large models, or when you need high-speed computations while keeping the same OS disk. You can choose any of the N series GPUs enabled virtual machine SKUs with DSVM. Note GPU enabled virtual machine SKUs aren't supported on Azure free accounts.
The Windows editions of the DSVM come preinstalled with GPU drivers, frameworks, and GPU versions of deep learning frameworks. On the Linux editions, deep learning on GPUs is enabled on the Ubuntu DSVMs.
You can also deploy the Ubuntu or Windows editions of the DSVM to an Azure virtual machine that isn't based on GPUs. In this case, all the deep learning frameworks falls back to the CPU mode.
Data science training and education
Enterprise trainers and educators who teach data science classes usually provide a virtual machine image. The image ensures students have a consistent setup and that the samples work predictably.
The DSVM creates an on-demand environment with a consistent setup that eases the support and incompatibility challenges. Cases where these environments need to be built frequently, especially for shorter training classes, benefit substantially.
What's included on the DSVM?
See a full list of tools on both the Windows and Linux DSVMs here.
Learn more with these articles: