Share via

Compute Resource for ML

Benjamin Toubøl 0 Reputation points
2023-08-29T07:42:14.7766667+00:00

Hello,

I am new to azure and wanted to make a small testing env using the 12 months of free services, I can't really see anything about free compute services so I assume I have to pay for that and wanted to ask a question in regards to what compute service would be best suited for ML test env, I wanted to keep it cheap yet still viable for training on small to medium datasets.

To keep it cheap I decided to stick with a CPU, I've attached a picture of the options and wanted to ask if there was any reason not to choose Standard F4 v2, as it has more cores and is cheaper than the general purpose DS3?

bill2

Azure Machine Learning
Azure Machine Learning
An Azure machine learning service for building and deploying models.
3,341 questions
Azure Virtual Machines
Azure Virtual Machines
An Azure service that is used to provision Windows and Linux virtual machines.
9,044 questions
0 comments

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. Ramr-msft 17,826 Reputation points
    2023-08-29T16:17:19.34+00:00

    @Benjamin Toubøl Thanks for the question, When choosing a compute service for your ML workload, it's important to consider the following factors:

    • The size of your dataset
    • The complexity of your model
    • The amount of time required to train your model
    • The level of parallelism required for your workload
    • The cost of the compute service

    Based on your requirements, a CPU-based compute service would be a good choice for your ML test environment. The Standard F4 v2 compute service is a good option for small to medium datasets, as it has 4 vCPUs and 8 GB of memory. The general-purpose DS3 compute service, on the other hand, has 4 vCPUs and 14 GB of memory, making it a good option for larger datasets.

    It's important to note that the cost of the compute service will depend on the size and duration of your workload. You can use the Azure pricing calculator to estimate the cost of each compute service based on your workload requirements.

    1 person found this answer helpful.
    0 comments

Your answer

Answers can be marked as Accepted Answers by the question author, which helps users to know the answer solved the author's problem.