ArcGIS Pro provides an integrated platform for 2D and 3D visualization and analysis. Pro has undergone extensive performance and scalability testing, especially in virtualized environments. Especially using the NVIDIA K1 and K2 graphics cards designed for virtualized environments. To share this information with our users, we will be publishing a series of blogs on the use of ArcGIS Pro in Citrix XenApp and XenDesktop, Microsoft Hyper-V VDI and VMWare Horizon View.
The Pro hardware requirements are being finalized. For a physical machine Pro does not require a GPU, but it is preferred. What about a virtual environment? Can a virtual environment provide access to a GPU? For a virtual environment the machine hosting the hypervisor can be configured with graphics cards specifically designed for virtualization, with the GPUs shared between multiple VMs. You may ask what type of user experience is possible in virtual environment and how many VMs can a single shared GPU accommodate? This and following blogs will provide information and answers to help you plan your Pro deployment in a virtualized environment.
These tests explore the use of virtualized GPUs in virtualized environments, and also share metrics that show when and how NVIDIA K1 and K2 graphics cards benefit the performance of ArcGIS Pro. The K1 and K2 cards are specifically designed for use in virtualized environments. Additional information on the NVIDIA K1 and K2 cards can be found at: http://www.nvidia.com/object/grid-boards.html
XenServer 6.2 and XenDesktop7.1
The first discussion is of using Pro in a virtualized environment using NVIDIA K1 cards, with Citrix XenServer 6.2 and XenDesktop 7.1. The K1 card was chosen for the first set of tests since it can support a higher number of concurrent VMs. Tests using the K2 card will be discussed in future blogs. The K1 card has 4 Kepler GPUs, each equivalent to a Quadro K600 card. Each physical GPU can host several types of virtual GPU (vGPU). An excellent source of information on this configuration and a description of the different vGPU types is at: http://www.citrix.com/content/dam/citrix/en_us/documents/go/reviewers-guide-remote-3d-graphics-apps-part-3-xenserver-vgpu.pdf
For the K1 card there are 3 vGPU types (K100, K140Q, and Pass-Through). Using Citrix XenCenter the hypervisor was configured to use the GPUs as the K140Q type. This virtual GPU type provides 1GB of video RAM to each vGPU. The K140Q configuration is the mid-tier configuration.
The K100 vGPU type is the lower capability configuration, suitable for business graphics, and can host up to 8 VMs per physical GPU (high-density). This vGPU type was tested and was found to provide a marginal user experience for 2D, but not adequate for 3D.
The Pass-through vGPU type is the highest capability configuration where a single VM is paired with a single GPU. When this was tested the 2D and 3D user experience was fantastic, with greater than 40 frames-per-second (FPS) reached for a majority of the test. This vGPU type have very low density, one VM to 1 GPU. This vGPU type will be probably be used when Pro is used for heavy-duty image analysis, 3D anaylsis, or when other high-end graphics applications require a dedicated GPU.
The K140Q vGPU type is the mid-tier configuration. This vGPU type provides an adequate user experience while Pro is using 3D, and a reasonable density at 4 VMs per GPU. The FPS were 13-17 which allowed for a relatively smooth user experience.
The next blog will dive into greater detail on the K140Q vGPU testing.