NVIDIA Virtual GPU (vGPU) brings the full benefit of NVIDIA hardware-accelerated graphics to virtualized solutions. This technology provides exceptional graphics performance for virtual desktops equivalent to local computers when sharing a GPU among multiple users. The following figures show how NVIDIA vGPU technology was used within this reference architecture. When sharing the compute and graphics resources, a single RTX 8000 GPU was used with 2GB vGPU framebuffers to support up to 24 users. The AI Guest VM on each VxRail in the cluster was assigned two vGPUs each with 48 GB of framebuffer that we mapped to the remaining two RTX 8000 GPUs within the VxRail.
The following figure shows a shared VDI and AI use case:
The following figure shows a VDI only use case:
The following figure shows an AI only use case:
NVIDIA vGPU is the industry's most advanced technology for sharing true GPU hardware acceleration between multiple virtual desktops without compromising the graphics experience.
NVIDIA vGPU offers multiple software variants to enable graphics for different virtualization techniques. This reference architecture uses the NVIDIA Quadro Virtual DataCenter Workstation (Quadro vDWS) software variant. Quadro vDWS is designed to provide workstation-grade performance in a virtual environment with support for up to four 4K or 5K monitors or up to two 8K monitors. We used this software to support a VDI workload with the appropriately sized frame buffer and features, and an AI workload that could take advantage of CUDA. CUDA is a parallel computing platform and programming model developed by NVIDIA that enables dramatic increases in computing performance by harnessing the power of GPUs.
Additional variants that are not used in this reference architecture are:
- GRID Virtual Applications—Designed to deliver graphics-accelerated applications using Remote Desktop Session Host (RDSH)
- GRID Virtual PC—Designed to provide full virtual desktops with up to dual 4K monitor support or single 5K monitor support
- NVIDIA Virtual Compute Server (vCS)—Designed to accelerate server virtualization so that the most compute-intensive workloads, such as artificial intelligence, deep learning, and data science, can be run in a VM
This reference architecture was configured with the following NVIDIA GPUs:
- NVIDIA RTX 8000—Select the Turing-based RTX 8000 for virtualized graphics performance for professional graphics and rendering workloads. The RTX 8000 has 48 GB of graphics frame buffer per card. Add up to three RTX 8000 GPU cards into your VxRail V570F to enable up 144 GB of frame buffer.
For additional GPU options, see the VxRail design and validation guides on the VDI Info Hub.