NVIDIA vGPU 19.0 Enables Graphics and AI Virtualization on NVIDIA Blackwell GPUs

TL;DR

• NVIDIA vGPU 19.0 expands graphics and AI virtualization on NVIDIA RTX PRO Blackwell GPUs, harnessing MIG-enabled capabilities for multi-tenant workloads.
• The RTX PRO 6000 Blackwell Server Edition ships with 96 GB of GDDR7 memory and supports up to 48 concurrent virtual machines (VMs) on a single GPU when MIG is enabled.
• A new 3B profile for NVIDIA Virtual PC (vPC) improves user experience for memory-intensive modern applications, boosting scalability and user density.
• NVIDIA AI Virtual Workstation (vWS) Toolkits deliver deployment guides and sizing for virtualized AI development, with a new Building an Agentic RAG toolkit for retrieval-augmented generation.
• Virtualization-Based Security (VBS) expands protection to Azure Local and Windows Server hypervisors, while EC2 G6f instances offer fractional vGPU options.
• Integration with Login VSI via NVIDIA nVector enables automated performance testing for GPU-enabled VDI environments.
• The release underscores NVIDIA’s focus on high-density, secure, and AI-enabled virtualized workloads in data centers. NVIDIA vGPU 19.0 release.

Context and background

Virtualization has long promised efficiency and scalability, yet rising demands from graphics and compute workloads challenge data centers seeking higher user density and cost efficiency. The NVIDIA RTX PRO 6000 Blackwell Series introduces the first NVIDIA Multi-Instance GPU (MIG) enabled GPUs designed to accelerate both graphics and compute workloads. The vGPU 19.0 release leverages the capabilities of the RTX PRO 6000 Blackwell Server Edition GPUs and NVIDIA RTX PRO Servers to improve performance across virtualized desktops and AI workloads, delivering stronger ROI for modern data centers. NVIDIA MIG technology partitions a single GPU into multiple isolated instances, each with its own memory, cache, engines, and streaming multiprocessors. This enables tighter QoS and uninterrupted execution for distinct workloads while sharing a single physical GPU. When paired with NVIDIA vGPU time-sliced sharing, MIG enables multi-tenancy within each partition, supporting diverse workloads on a single device. With MIG enabled, vGPU 19.0 supports up to 48 concurrent VMs on a single GPU, enabling use cases ranging from business operations and graphics rendering to AI development and multimodal inference. NVIDIA vGPU 19.0 release. Windows and AI workloads continue to push memory and graphics requirements higher. Internal testing noted that GPU memory usage for knowledge worker applications on Windows 11 rose by about 60% compared to Windows 10, highlighting the need for robust virtualization profiles and scalable configurations. In response, vGPU 19.0 adds a new 3B profile for vPC to balance modern accelerated graphics needs with scalable user density. NVIDIA vGPU 19.0 release. NVIDIA is expanding tooling for AI development in virtualized environments. The NVIDIA AI Virtual Workstation (vWS) Toolkits provide deployment guides and sizing recommendations tailored to existing virtualized infrastructures, helping IT teams plan for AI workloads. A new Building an Agentic RAG toolkit focuses on building AI agents that use retrieval-augmented generation to fetch information from documents and the web, improving response quality in AI-enabled workflows. AI vWS demo videos illustrate these capabilities. NVIDIA vGPU 19.0 release. Security in virtualization is strengthened through Virtualization-Based Security (VBS), which uses hardware virtualization and hypervisors to create isolated environments separate from the OS. vGPU 19.0 adds VBS support with Microsoft Azure Local and Microsoft Windows Server hypervisors, delivering enhanced protection for regulated sectors such as financial services, healthcare, and government. NVIDIA vGPU 19.0 release. For cloud and edge deployments, Amazon EC2 G6f instances accelerated by NVIDIA L4 Tensor Core GPUs offer fractional vGPU options. These instances provide configurations from one-eighth to one-half of a GPU, enabling cost-effective scaling for workloads like NLP, graphics, and game streaming. NVIDIA vGPU 19.0 release. The NVIDIA nVector benchmarking tool continues to help simulate knowledge worker workflows at scale, with recent integration with Login Enterprise enabling automated testing for GPU-enabled VDI—assessing graphical responsiveness and CPU offloading for mainstream applications and providing detailed performance insights. NVIDIA vGPU 19.0 release.

What’s new

The vGPU 19.0 release introduces several notable updates for Blackwell-based systems:

• 3B profile for NVIDIA Virtual PC (vPC) to improve user experience for modern, graphics-heavy applications while maintaining scalability and density.
• NVIDIA AI Virtual Workstation (vWS) Toolkits with detailed deployment guides and sizing for existing virtualized environments, plus a Building an Agentic RAG toolkit for retrieval-augmented generation.
• Virtualization-Based Security (VBS) support with Microsoft Azure Local and Microsoft Windows Server hypervisors to strengthen security in virtualized workloads.
• Support for AWS-based EC2 G6f instances powered by NVIDIA L4 Tensor Core GPUs, offering fractional vGPU configurations from 3 GB to 12 GB.
• Integration with NVIDIA nVector and Login VSI for automated testing and performance analytics in GPU-enabled VDI environments.
• Optimizations to unlock up to 48 concurrent VMs per GPU on MIG-enabled RTX PRO Blackwell GPUs, enabling higher density across diverse workloads. NVIDIA vGPU 19.0 release.

Why it matters (impact for developers/enterprises)

For enterprises, vGPU 19.0 represents a strategic move toward higher user density without sacrificing performance or security. MIG-enabled GPUs partition a single device into multiple isolated instances, delivering predictable QoS for each workload and reducing the total number of GPUs required to support modern virtual desktops, graphics rendering, and AI workloads. The 3B vPC profile helps teams run memory-intensive apps more smoothly, expanding the range of viable workloads in virtualized data centers. Security is enhanced through VBS, which isolates sensitive processes and data from the host OS, a key factor for regulated industries adopting virtualized infrastructures. The Azure Local and Windows Server hypervisor support broadens deployment options for enterprise customers with hybrid cloud strategies. In cloud environments, AWS EC2 G6f with fractional vGPU capacity broadens the range of resource configurations, enabling cost-efficient scaling for AI-powered NLP, graphics tasks, and game streaming. From a developer perspective, the new vWS Toolkits and the Agentic RAG toolkit provide practical resources to accelerate AI workflow setup, experimentation, and production planning within virtualized environments. The integration with Login VSI and NVIDIA nVector equips teams with automated performance testing and scalability analytics, helping ensure consistent user experiences at scale. NVIDIA vGPU 19.0 release.

Technical details or Implementation

NVIDIA MIG technology enables spatial partitioning of a single GPU into multiple, isolated instances, each with dedicated memory, cache, engines, and streaming multiprocessors. When combined with NVIDIA vGPU’s time-sliced sharing, this architecture supports multi-tenancy within each MIG, allowing a single GPU to host diverse workloads concurrently. With MIG enabled on RTX PRO Blackwell GPUs, vGPU 19.0 supports as many as 48 VMs per GPU, illustrating the potential for dense virtualized infrastructure. The RTX PRO 6000 Blackwell Server Edition brings 96 GB of ultra-fast GDDR7 memory to bear on demanding enterprise workloads, from multimodal AI inference and scientific computing to graphics and video applications. This hardware baseline, together with the software stack, underpins improved performance and ROI for virtualized data centers. A notable performance-related observation from internal testing highlights the memory demands of modern Windows environments. GPU memory usage for knowledge worker applications on Windows 11 was observed to be 60% higher than on Windows 10, motivating the adoption of richer profiles like the 3B vPC to maintain user density without compromising experience. NVIDIA vGPU 19.0 release. The vWS Toolkits streamline AI development by offering deployment guides and sizing guidance tailored for existing virtualized environments. The Building an Agentic RAG toolkit provides a framework for constructing AI agents that leverage retrieval-augmented generation for dynamic information retrieval from documents and web searches, improving answer relevance in AI-powered applications. Demonstrations and training materials are available via AI vWS demo videos. NVIDIA vGPU 19.0 release. Security-conscious deployments benefit from VBS support with Azure Local and Windows Server hypervisors, enabling isolated execution environments even when an OS is compromised. This capability complements the broadening support across cloud and on-premises hypervisor options. NVIDIA vGPU 19.0 release. Cloud-ready scalability is reinforced by AWS EC2 G6f support, offering fractional vGPU configurations on L4 Tensor Core GPUs. This enables on-demand resource provisioning aligned to workload needs, including NLP, graphics, and game streaming. NVIDIA vGPU 19.0 release. NVIDIA nVector continues to serve as a benchmarking tool for knowledge worker workflows, with integration to Login Enterprise for automated performance testing of GPU-enabled images, applications, and desktops. This joint capability provides performance analysis to ensure a consistent user experience at scale. NVIDIA vGPU 19.0 release.

Key takeaways

• MIG-enabled RTX PRO Blackwell GPUs deliver high-density virtualization with up to 48 VMs per GPU.
• A new 3B vPC profile supports modern, graphics-heavy applications while preserving scalability.
• VBS support broadens security coverage for Azure Local and Windows Server hypervisors.
• AI development tooling (vWS Toolkits, Agentic RAG) accelerates virtualized AI workflows.
• Fractional vGPU options on AWS EC2 G6f expand cost-effective deployment options.
• Integration with Login VSI and nVector enables automated testing and performance insights for GPU-enabled VDI environments.
• The RTX PRO 6000 Blackwell Server Edition’s 96 GB GDDR7 memory helps address demanding enterprise workloads. NVIDIA vGPU 19.0 release.

References

• NVIDIA vGPU 19.0 release: https://developer.nvidia.com/blog/nvidia-vgpu-19-0-enables-graphics-and-ai-virtualization-on-nVIDIA-blackwell-gpus