About the A8, A9, A10, and A11 compute-intensive instances
Here is background information and some considerations for using the Azure A8, A9, A10, and A11 instances, also known as compute-intensive instances. This article focuses on using these instances for Windows VMs. This article is also available for Linux VMs.
Azure has two different deployment models for creating and working with resources: Resource Manager and classic. This article covers using both models, but Microsoft recommends that most new deployments use the Resource Manager model.
Key features of these instances include:
High-performance hardware – The Azure datacenter hardware that runs these instances is designed and optimized for compute-intensive and network-intensive applications, including high-performance computing (HPC) and batch applications, modeling, and large-scale simulations.
RDMA network connection for MPI applications – Set up the A8 and A9 instances to communicate with other A8 and A9 instances over a low-latency, high-throughput network in Azure that is based on remote direct memory access (RDMA) technology. This feature can boost the performance of certain Linux and Windows Message Passing Interface (MPI) applications.
Support for HPC clusters – Deploy cluster management and job scheduling software on the A8, A9, A10, and A11 instances in Azure to create a stand-alone HPC cluster or to add capacity to an on-premises cluster.
A10 and A11 instances have the same performance optimizations and specifications as the A8 and A9 instances. However, they do not include access to the RDMA network in Azure. They are designed for HPC applications that do not require constant and low-latency communication between nodes, also known as parametric or embarrassingly parallel applications.
The Azure A8, A9, A10, and A11 compute-intensive instances feature high-speed, multicore CPUs and large amounts of memory, as shown in the following table.
|A8 and A10||Intel Xeon E5-2670 |
8 cores @ 2.6 GHz
|DDR3-1600 MHz |
|A9 and A11||Intel Xeon E5-2670 |
16 cores @ 2.6 GHz
|DDR3-1600 MHz |
Additional processor details, including supported instruction set extensions, are at the Intel.com website.
A8 and A9 instances have two network adapters, which connect to the following two back-end Azure networks.
|10-Gbps Ethernet||Connects to Azure services (such as Azure Storage and Azure Virtual Network) and to the Internet.|
|32-Gbps back end, RDMA capable||Enables low-latency, high-throughput application communication between instances within a single cloud service or availability set. Reserved for MPI traffic only.|
See Access to the RDMA network in this article for additional requirements for MPI applications to access the RDMA network.
A10 and A11 instances have a single, 10-Gbps Ethernet network adapter that connects to Azure services and the Internet.
Azure subscription – If you want to deploy more than a small number of compute-intensive instances, consider a pay-as-you-go subscription or other purchase options. If you're using an Azure free account, you can use only a limited number of Azure compute cores.
Cores quota – You might need to increase the cores quota in your Azure subscription from the default of 20 cores per subscription (if you use the classic deployment model) or 20 cores per region (if you use the Resource Manager deployment model). To request a quota increase, open a support ticket at no charge as shown in Understanding Azure limits and increases.
Virtual network – An Azure virtual network is not required to use the compute-intensive instances. However, you may need at least a cloud-based Azure virtual network for many scenarios, or a site-to-site connection if you need to access on-premises resources such as an application license server. You will need to create a new virtual network to deploy the instances. Adding an A8, A9, A10, or A11 VM to a virtual network in an affinity group is not supported.
Cloud service or availability set – To connect through the RDMA network, size A8 or A9 VMs must be deployed in the same cloud service (if you use the classic deployment model) or the same availability set (if you use the Azure Resource Manager deployment model).
Pricing - The A8–A11 VM sizes are available only in the Standard pricing tier.
Resizing – You can't resize an instance of size other than A8–A11 to one of the compute-intensive instance sizes (A8-11), and you can’t resize a compute-intensive instance to a non-compute-intensive size. This is because of the specialized hardware and the performance optimizations that are specific to compute-intensive instances.
RDMA network address space - The RDMA network in Azure reserves the address space 172.16.0.0/16. If you plan to run MPI applications on A8 and A9 instances in an Azure virtual network, make sure that the virtual network address space does not overlap the RDMA network.
Within a single cloud service, availability set, or Azure Batch pool, the A8 and A9 instances can access the RDMA network in Azure to run Windows MPI applications that use the Microsoft Network Direct interface to communicate between instances.
See the following table for prerequisites for MPI applications to access the RDMA network in Windows virtual machines, cloud services, and Azure Batch pools of the A8 or A9 instances. For typical deployment scenarios, see Set up a Windows RDMA cluster with HPC Pack to run MPI applications and Use multi-instance tasks to run Message Passing Interface (MPI) applications in Azure Batch.
|Prerequisite||Virtual machines||Cloud services or Batch pool|
|Operating system||Windows Server 2012 R2 or Windows Server 2012||Windows Server 2012 R2, Windows Server 2012, or Windows Server 2008 R2 Guest OS family|
|MPI||MS-MPI 2012 R2 or later, or Intel MPI Library 5||MS-MPI 2012 R2 or later, or Intel MPI Library 5|
On size A8 and A9 virtual machines, the HpcVmDrivers extension must be added to the VMs to install Windows network device drivers that are needed for RDMA connectivity. Depending on your deployment method, the HpcVmDrivers extension might be added to a size A8 or A9 VM automatically, or you might need to add it yourself. To add the extension yourself, see Manage VM extensions.
HPC Pack is not required for you to use the A8, A9, A10, and A11 instances with Windows Server, but it can be a useful tool to run Windows-based MPI applications that access the RDMA network in Azure. HPC Pack 2012 R2 and later versions include a runtime environment for the Microsoft implementation of the Message Passing Interface (MS-MPI) for Windows that can use the Azure RDMA network when deployed on A8 and A9 instances.
For more information and checklists to use the compute-intensive instances with HPC Pack on Windows Server, see Set up a Windows RDMA cluster with HPC Pack to run MPI applications.
For storage capacities and disk details, see Sizes for virtual machines.
To get started deploying and using A8 and A9 instances with HPC Pack on Windows, see Set up a Windows RDMA cluster with HPC Pack to run MPI applications.
For information about using A8 and A9 instances to run MPI applications with Azure Batch, see Use multi-instance tasks to run Message Passing Interface (MPI) applications in Azure Batch.