Introducing NVv4 Azure Virtual Machines for GPU visualization workloads

Publicado em 8 agosto, 2019

Senior Program Manager, Microsoft Azure

Azure offers a wide variety of virtual machine (VM) sizes tailored to meet diverse customer needs. Our NV size family has been optimized for GPU-powered visualization workloads, such as CAD, gaming, and simulation. Today, our customers are using these VMs to power remote visualization services and virtual desktops in the cloud. While our existing NV size VMs work great to run graphics heavy visualization workloads, a common piece of feedback we receive from our customers is that for entry-level desktops in the cloud, only a fraction of the GPU resources is needed. Currently, the smallest sized GPU VM comes with one full GPU and more vCPU/RAM than a knowledge worker desktop requires in the cloud. For some customers, this is not a cost-effective configuration for entry-level scenarios.

Announcing NVv4 Azure Virtual Machines based on AMD EPYC 7002 processors and virtualized Radeon MI25 GPU.

The new NVv4 virtual machine series will be available for preview in the fall. NVv4 offers unprecedented GPU resourcing flexibility, giving customers more choice than ever before. Customers can select from VMs with a whole GPU all the way down to 1/8th of a GPU. This makes entry-level and low-intensity GPU workloads more cost-effective than ever before, while still giving customers the option to scale up to powerful full-GPU processing power.

NVv4 Virtual Machines support up to 32 vCPUs, 112GB of RAM, and 16 GB of GPU memory.

 

Size vCPU Memory GPU memory Azure network
Standard_NV4as_v4 4 14 GB 2 GB 50 Gbps
Standard_NV8as_v4 8 28 GB 4 GB 50 Gbps
Standard_NV16as_v4 16 56 GB 8 GB 50 Gbps
Standard_NV32as_v4 32 112 GB 16 GB 50 Gbps

With our hardware-based GPU virtualization solution built on top of AMD MxGPU and industry standard SR-IOV technology, customers can securely run workloads on virtual GPUs with dedicated GPU frame buffer. The new NVv4 Virtual Machines will also support Azure Premium SSD disks. NVv4 will have simultaneous multithreading (SMT) enabled for applications that can take advantage of additional vCPUs.

For customers looking to utilize GPU-powered VMs as part of the desktop as a service (DaaS) offering, Windows Virtual Desktop provides a comprehensive desktop and application virtualization service running in Azure. The new NVv4-series Virtual Machines will be supported by Windows Virtual Desktop as well as Azure Batch  for cloud-native batch processing.

Remote display application and protocols are key to a good end user experience with VDI/DaaS in the cloud. The new virtual machine series will work with Windows Remote Desktop (RDP) 10, Teradici PCoIP, and HDX 3D Pro. The AMD Radeon GPUs support DirectX 9 through 12, OpenGL 4.6, and Vulkan 1.1.

Customers can sign up for NVv4 access today by filling out this form. NVv4 Virtual Machines will initially be available later this year in the South Central US and West Europe Azure regions and will be available in additional regions soon thereafter.