Clarification needed - Pricing Comparison B/w Machine Learning studio (Classic) and Azure Machine Learning

Question

Karthik Dulam on Tue, 28 Jan 2020 21:06:40

Hello Team,

I have been using the Azure Machine Learning Studio(Classic) for deploying webservices (Models created locally in R). We have been paying 9.95\$ for Studio access and 100.13\$ for Standard tier Webservices plan.

I have read about the new Azure Machine Learning service and would want to understand more about the costs involved in that. Here is what i came up with based on my understanding. Can some one correct me in case i am wrong in my analysis?

• The Studio (Classic) service costs us \$9.99 per workspace per month and \$1 per hour of experiment usage. So assuming 20 hours of experiment usage per month, it would come up to \$40 per month as we have two active workspaces. Apart from this, we are also charged for web API pricing. We are on standard S1 tier and we are paying \$100.13 per month. So total resource costs are around \$140.13 per month.
• The newer service is coming with compute charges included. Please correct me if i wrong in assuming that i will not be paying the web API pricing fee (S1 Tier) as above once i transition to the newer service as i assume that the compute charges are included in the calculation below.
• i.e. Suppose I train a model for 100 hours using a workspace in US East and then deploy it to be available for all 30 (730 hours) days in a month.
• Training
• Azure VM Charge: (1 machines * \$0.097 per machine) * 100 hours (1month) = \$9.7
• Azure Machine Learning Charge: (1 machines * 16 cores * \$0 per core) * 10 hours = \$0
• Total: \$9.7 + \$0 = \$9.7
• Deploy
• Azure VM Charge: (1 machines * \$0.097 per machine) * 730 hours (1month) = \$70.81
• Azure Machine Learning Charge: (1 machines * 16 cores * \$0 per core) * 730 hours = \$0
• Total: \$70.81 + \$0 = \$70.81

Total Charges for new service : \$9.7 (training) + \$70.81 (Deployment) = \$80.51

Am i correct to assume that for the use case presented, the charges are correct or are there any additional charges that i would incur by going with the new machine learning service.

Replies

GiftA-MSFT on Tue, 28 Jan 2020 23:06:21

Hi,

We are reviewing this information and will provide feedback as soon as possible. Thanks.

Regards,

GiftA-MSFT on Wed, 29 Jan 2020 20:36:07

Hi,

I suggest you use Azure Pricing Calculator to get accurate estimates. Select 'Virtual Machines' from the 'Compute' tab, select 'Azure Machine Learning' and 'Azure Machine Learning Studio' from the 'AI+Machine Learning' tab. The services you’ve selected should show up as you scroll down the page. You can then use the interactive tool to generate your estimate. For the new service, free API usage cost only applies to Dev/Test tier and includes 1,000 API transactions and 2 API compute hours. Hope this helps. Thanks.

Regards,

Karthik Dulam on Wed, 29 Jan 2020 20:59:39

Adding Azure Machine Learning in the Azure Pricing Calculator includes the charge for the Virtual Machine. Am i correct? If yes, then why do we need to estimate price for Virtual machines again ?

Please ignore if i sound dumb :) This is something new for me and trying to figure out what is the best way forward without adding additional overhead costs.

Also, are there any plans to discontinue Machine learning studio (Classic) in future?

RohitMungi-MSFT on Thu, 30 Jan 2020 14:26:11

Hello Karthik,

Yes, the pricing calculator displays Azure VM when you add a Machine learning workspace. You can use the same compute target for all experiments but practically users use additional compute targets in their workspace which can be estimated now by adding Azure VM to the pricing calculator.

Classic is the legacy ML studio of Azure with rich functionality and services for Machine Learning. But, Customers currently using or evaluating Machine Learning Studio (classic) are encouraged to try Azure Machine Learning designer (preview), which provides drag-n-drop ML modules plus scalability, version control, different compute targets and enterprise security where latest releases are published.

-----------------------------------------------------------------------------------------------------------