Azure Machine Learning cost

Question

Hello,

I am new to Azure Machine Learning, I have some questions related to cost

1) What will be the cost breakup? (Workspace charges, Network charges, Disk Charges etc.)

2) I saw that there were some charges for Bandwidth and Load balancer. Why is it getting charged and I am not able to see the details in Azure ML?

3) I have a SQL Sever which is inside a Vnet (VPN Gateway) and I want to integrate it with Azure Machine Learning and use the data for my analysis. What all will be the charges for it?

4) Is there a way to stop the computes automatically when i am not using it, as i dont want to be charged when i am not using the Azure Machine learning?

5) Will there be any charges even though i am not using the Azure Machine learning? (Static Charges)

Answer 1

@Srinivasan G Here are the charges that you could incur for the above setup:

1) What will be the cost breakup? (Workspace charges, Network charges, Disk Charges etc.)

For Azure Machine learning the charges of ML services are nil for enterprise edition workspaces. Currently all the workspaces are using enterprise edition and if there are any basic edition workspaces they can be migrated with no down time. All other charges while using Azure ML workspace depends on the compute used and the setup of the compute. If they are enabled in a vnet then you could incur data charges according to vnet's pricing.

2) I saw that there were some charges for Bandwidth and Load balancer. Why is it getting charged and I am not able to see the details in Azure ML?

The charges incurred for LB and bandwidth could be based on your setup and how the compute was setup to run the experiments. If you have also setup your designer to use virtual network then there could be charges on how the compute was setup with respect to the region. More details of the setup of workspace with private networks are detailed here. For the breakup of charges mentioned above you can raise a support request through the azure portal for billing which does not require any support plan to raise a ticket from the Help+Support tab on Azure portal.

3) I have a SQL Sever which is inside a Vnet (VPN Gateway) and I want to integrate it with Azure Machine Learning and use the data for my analysis. What all will be the charges for it?

Once you are able to confirm if your vnet connection is successful you can get the data from Azure SQL database using the import data module from Azure ML designer. If you are planning to setup your own SQL server then this functionality is not supported with the available options, you might need to use the option URL via HTTP to get the required data in your experiments.

If you are attaching storage from different region than workspace region, it can result in higher latency and additional network usage costs.

4) Is there a way to stop the computes automatically when i am not using it, as i dont want to be charged when i am not using the Azure Machine learning?

The option to shutdown compute instance automatically that is not in use is not available currently. You can stop the compute instances that are not required, for compute clusters you can set the minimum no. of nodes to 0 to ensure no compute is running when not in use.

5) Will there be any charges even though i am not using the Azure Machine learning? (Static Charges)

Azure ML learning enterprise edition workspaces do not have a surcharge i.e charges for using Azure ML. If there are any associated compute or storage or virtual networks there could be charges if used.

Answer 2

I had the same issue recently.
I was being charged money for "load balancer" while obviously there is no place in Azure ML to start or stop a load balancer. I have checked my ML Studio under 'Compute' section make sure there is no running compute instances or clusters in ML.

How I solved the issue:
I realised that I have created a real-time inference endpoint and I suspect that can be costing me money. I deleted that endpoint and in the next few days I can see the 'load balancer' cost has disappeared. Problem solved.

Conclusion:
a real-time inference endpoint requires some computing resources (I believe it's an AKS cluster in this case) so there should be a cost.
However, this computing resource is not shown under the 'compute' section. Not under any of the 4 tabs there (compute instances, computer clusters, inference clusters, attached computes). I believe it should show under the "inference cluster" tab in the compute section, but it didn't.
In 'cost analysis', this cost has been incorrectly categorised to "load balancer". This categorisation is not only very confusing but also incorrect.

@romungi-MSFT Could you please advise if my finding is correct or not?

Answer 3

@Tao
I too saw a lot of Load Balancer charged in my invoice, as a matter of fact I got charged everyday for Load Balancer for 2 months. I am sure that the total amount of time I would have kept a Real Time Endpoint live would not exceed 3-4 hours during the course of 2 months. I have contacted helpdesk, looking forward to their response.

Answer 4

Hi,
I also noticed some load balancer costs today and found a way to stop the LB costs.
Based on this Microsoft Learn: https://github.com/MicrosoftDocs/azure-docs/blob/main/articles/machine-learning/concept-plan-manage-cost.md#costs-might-accrue-before-resource-deletion

Each VM is billed per hour it is running. Cost depends on VM specifications. VMs that are running but not actively working on a dataset will still be charged via the load balancer. For each compute instance, one load balancer will be billed per day. Every 50 nodes of a compute cluster will have one standard load balancer billed. Each load balancer is billed around $0.33/day. To avoid load balancer costs on stopped compute instances and compute clusters, delete the compute resource. One virtual network will be billed per subscription and per region. Virtual networks cannot span regions or subscriptions. Setting up private endpoints in vNet setups may also incur charges. Bandwidth is charged by usage; the more data transferred, the more you are charged.

Answer 5

@Mohammad Bakharzy

I had created my VMs on Jan 2021. My billings were close to Rs 800 per month($12 per month). Not only that, it remained consistent and only changed in a minor way whenever I used them for any ML training work.

But now all of a sudden since Feb 2022 I am being charged for a LB! and my costs have gone up >3x. I've been in touch with some people in the billing team but they are taking their own sweet time to respond.

I am willing to accept that this is a fair charge as long as they can answer the following:

Locate the LB linked to my account
Is the LB charge a new addition ? if not then why was I never billed before for instances I've owned since Jan 2021 ?