Creating Ampere AI Virtual Machines on Microsoft Azure
Tutorial
Introduction
Users don’t have to wait until the general availability of Ampere Virtual Machines on Azure Marketplace to run their AI inference workloads, Ampere Altra resources are already available on Microsoft Azure Community Images in Preview. In this document we describe the step-by-step process for instantiating an Ampere Altra compute resource on Microsoft Azure for running Ampere optimized TensorFlow for AI inferencing. The recommended resource type is either Dplsv5 or Dpldsv5 depending on whether the application requires local storage or not.
Step 1
-
Go to azure.microsoft.com and sign into your account.
-
Note that you will need a paying account to create and run these resources.
-
Select “Virtual machines”, then select “Create a virtual machine”.
Step 2
Once on the virtual machines page select “Create” then select “Azure virtual machine”.
Step 3
On the next page select “See all images” to locate the available Ampere resources.
Step 4
Select “Community images (PREVIEW)".
Step 5
To quickly find the Ampere resources, type “Ampere” in the “Publisher name” field.
The “tensorflow_ampere_optimized_test” image will appear. PyTorch and ONNX-RT resources can also be accessed through the same process.
Selecting this resource will take you back to the previous screen.
You will notice that the Image Architecture button will be automatically set to Arm64. In addition, the system will have picked a 16 vCPUs with 64GB of memory. The usage rate for this resource (as of 12/1/22) is $449.68/month
Step 6
Name your Virtual machine. Once your virtual machine is named, Azure will generate an SSH key.
Step 7
Click “Review + Create” to review the configuration of your virtual machine and read the licensing agreements. Finalize the creation of the virtual machine by clicking on “Create” at the bottom left corner of the page.
Ampere Computing
4655 Great America Parkway
Suite 601 Santa Clara, CA 95054
