Back in 2019 when Azure launched the first GPU-partitioned (GPU-P) virtual machine (VM) offering in the public cloud, our customers loved it and asked for a similar offering on the edge with Azure Stack HCI. Our customers wanted the flexibility to choose the right-sized GPU acceleration and get the benefits of GPU-P, which enables effective GPU sharing, leading to cost-optimized and efficient configurations. While working with GPU manufacturers, we are launching our first implementation with NVIDIA.
In collaboration with NVIDIA, we are pleased to announce the general availability of GPU partitioning (GPU-P) on NVIDIA A2, A10, A16 and A40 GPUs in Azure Stack HCI, enabled with NVIDIA RTX Virtual Workstation (vWS) and NVIDIA Virtual PC (vPC) software, starting with version 22H2, with more GPUs coming soon. Expanded support for NVIDIA GPUs and NVIDIA RTX vWS in Azure Stack HCI, combined with GPU-P capabilities, enables technical and creative professionals to count on powerful virtual workstations, available from anywhere, that have the computational power required for data science and visual computing, including 3D design, photorealistic simulations, and stunning visual effects. Purpose-built for high-density, graphics-rich virtual desktop infrastructure (VDI), NVIDIA vPC software with the NVIDIA A16 GPU, provides double the user density versus the previous generation, while ensuring the best possible user experience for knowledge workers.
We are always innovating to provide hybrid infrastructure that makes it easy for customers to manage their workloads at the edge and in the Azure cloud. GPU-P is implemented using single root I/O virtualization (SR-IOV), which provides a strong, hardware-backed security boundary with predictable performance for each virtual machine. The GPU-P technology leverages NVIDIA GPUs and NVIDIA virtual GPU (vGPU) software such as NVIDIA RTX Virtual Workstation, NVIDIA Virtual PC, and NVIDIA Virtual Applications to right-size the GPU-powered virtual machines for the workload.
Azure Virtual Desktop for Azure Stack HCI also works with GPU partitioning on Azure Stack HCI. Customers can now scale their on-premises virtual desktop workloads by assigning different GPU partitions to their VDI VMs, thereby increasing the user density by an order of magnitude.
“GPU partitioning support for Azure Stack HCI is transformational for customers seeking cost-effective solutions at the edge that require GPU acceleration. With GPU-P, now customers can scale their VDI workloads as well as AI/ML inferencing scenarios in distributed locations like stores and factories,” said Timothy Isaacs, General Manager of Azure Edge and Infra.
“To meet the requirements of creative and technical professionals, IT teams must architect many components, including performance and user experience, sizing, and cost utilization,” said Bob Pette, vice president of Professional Visualization at NVIDIA. “Combining the complementary technologies of NVIDIA GPUs and NVIDIA RTX vWS with Microsoft Azure Stack HCI helps meet the diverse demands of users and their complex visualization workloads.”
With support for NVIDIA vGPU, customers can create up to 16 partitions and cost-effectively scale their workloads. Watch this demo to see how you can use Windows Admin Center to create and manage the GPU partitions.
Azure Stack HCI integrated systems are the fastest and easiest way to get up and running with Azure Stack HCI. Every integrated system is jointly supported by Microsoft and the solution partner to deliver a seamless appliance-like experience. Azure Stack HCI solutions that support GPU-P are available starting with DataON (AZS-6208G), Dell (AX-750), and Lenovo (Lenovo ThinkAgile MX3530), and many other Azure Stack HCI solutions partners soon.
For more details on GPU-P support on Azure Stack HCI, please visit our public documentation on GPU Partitioning.
Keep up to date on Azure’s collaboration and offerings with NVIDIA by reading the latest blog on the NVIDIA vGPU 15 announcement.
Find more information about NVIDIA RTX Virtual Workstation, NVIDIA Virtual PC, NVIDIA Virtual Applications here.
Additional links: NVIDIA Virtual GPU Software Documentation v15.0