azure machine learning teaching

Erdi Kara 1

Hello
I have 5000$ Azure grant that I will use to teach my deep learning class. We will have 5-6 groups with two students each. I have lecture notes and assignments in the form of Juptyter notebooks.
This is my first time I use Azure. I managed to train a model with default Tesla K80. It takes 6 mins to runs a model that I run in 18 seconds with my RTX-2070 Super in my local machine. My question what can be a best practice to teach this class. Compatring to Colab, I found Azure extremely confusing and difficult to navigate. Should I create multiple workspaces under a single subscription? If so, how can I fix that slow GPU issue, should I submit quota increase for each workspace?
Any kind of help would be highly appreciated.

romungi-MSFT 41,866 Reputation points Microsoft Employee

2023-02-02T12:44:50.12+00:00

Erdi Kara If you would like to have dedicated workspaces for each group or user then you can create multiple workspaces in a single subscription. However, creating a single workspace and using single compute instance or cluster can be economical if you each group works individually. They can basically reuse the compute and you can use name your experiments based on group to retrieve and reuse the jobs/models.

Slowness issue might also include compute setup time rather than the total time to run your experiment. If you plan to keep your compute running then subsequent runs should not be taking longer times. You can always request for quota increase through a support case if you are seeing persistent slowness with the compute.