As you may already be seeing lots of information flowing around in the internet around the GRID virtualization solution delivered by nVIDIA, there have been some quick and significant changes (improvements) over the last few years around the GRID virtualization offerings from nVIDIA GRID integration with VMware, Citrix and Microsoft for the Blast Extreme/PCoIP, HDX/3D Pro and RemoteFX protocol respectively, to deliver Graphics accelerated virtual apps and desktops. Per my opinion, out of the three leading vendors, VMware and Citrix have been aggressive in their commitment compared to Microsoft. However it may be aggressive with the introduction of Windows Server 2016 architecture and beyond.
Having said that, I thought why not just put together a quick reference of the three primary GRID Card offerings from nVIDIA, their use-cases, vGPU profiles, licensing and other requirements. This may come handy for the Sales and the Pre-Sales team to basically engage with their customer base to identify the needs of User Experience at its highest quality apart from the graphics intensive apps & desktop virtualization.
Important Note (Disclaimer): Please be aware while reading this that many of the features, functionalities may not have been tested, validated and thus may change from real-time results. Therefore, I strongly recommend you to leverage (as applicable) the official documentation, whitepaper and blogs from nVIDIA, Citrix, VMware and Microsoft or other reliable sources, as there are continuous changes and updates are being released constantly. Few of the important official documentation are listed in the references section at the bottom of this blog.
GRID Solution Matrix with VMware, Citrix and Microsoft
|GRID Software Editions|
|Features & Components||GRID Virtual Application (vApp)||GRID Virtual PC (vPC)||GRID Virtual WorkStation (vWS)|
|nVIDIA GPU cards supported||Tesla M10 (User
|Tesla M60 (Performance-Optimized)||Tesla M6 (Blade-
|Remoting App & Desktop solutions supported||XenApp, Horizon Hosted Apps (RDSH/ThinApp)||XenDesktop, Horizon View/vSGA, RemoteFX, vDGA||Workstation PCs with GPU passthrough, High-end VDI (desktop OS)|
|Use-cases or user profiles||Knowledge/task worker||Business/Power User/Medium 3D worker||High-end & powerful Designers / 3D app users|
|Client OS supported||N/A||Windows||Windows/Linux|
|Maximum Resolution||N/A||2560×1600||4096×2160 (4K)|
|CUDA & OpenCL Supported||Yes (only on 8GB 1:1 profiles i.e. M10-8A and M10-8Q)||Yes (only on 8GB 1:1 profiles i.e. M60-8A and M60-8Q)||Yes (only on 8GB 1:1 profiles i.e. M6-8A and M6-8Q)|
|DirectX 12, Direct2D, and DirectX Video Acceleration (DXVA)||Yes||Yes||Yes|
|NVIDIA GRID SDK (remote graphics acceleration)||Yes||Yes||Yes|
|GPU Pass-through Supported||Yes (only on 8GB 1:1 profiles)||Yes (only on 8GB 1:1 profiles)|
|Bare-Metal Supported||Yes (Only NVIDIA Tesla M6 Hardware supported as primary display device)||Yes (Only NVIDIA Tesla M6 Hardware supported as primary display device)|
|GRID Card Technical Specifications|
|GPU Card||Tesla M10||Tesla M60||Tesla M6|
|OEM hardware consideration||designed for rack and tower servers, optimized for maximum user density per host||designed for rack and tower servers, optimized for performance||designed for blade servers and converged|
|Number of GPUs||Quad Mid-Level Maxwell||Dual High-End Maxwell||Single High-End Maxwell|
|Total NVIDIA CUDA®Cores||2,560 (640 per GPU)||4,096 (2,048 per GPU)||1,536|
|Total Memory Size||32 GB GDDR5 (8 GB per GPU)||16 GB GDDR5 (8 GB per GPU)||8 GB GDDR5|
|Max vGPU Instances||64||32||16|
|Max Power||225 W||300 W||100 W|
|Form Factor||PCIe 3.0 Dual Slot (rack)||PCIe 3.0 Dual Slot (rack)||MXM (blade)|
|Board Dimensions||10.5″ x 4.4″||10.5″ x 4.4″||3.2″ x 4.1″|
|Cooling Solution||Passive||Passive / Active||Bare Board|
|GRID Licensing Model – Concurrent User (CCU)|
|Licence Type – Option 1||Perpetual (one-time) + SUMS* (first year mandatory)|
|Licence Type – Option 2||Annual Subscription – pay as you go (yearly renewal)|
|License Entitlement||vApps||vPC + vApps (mixed environment)||vWS + vApps (mixed environment)|
|*Support, Updates, and Maintenance Subscriptions (SUMS) ensures that you have 24×7 access to technical support, along with timely software patches, updates, and upgrades. SUMS is included in your NVIDIA GRID software subscription, but is a required one-year add-on if you choose a perpetual license. NVIDIA GRID K1 and K2 GPUs do not require a license to run vGPU.|
|License Bundle Inclusions|
|GPU Card||Tesla M10||Tesla M60||Tesla M6|
|Graphics drivers – Windows||√|
|Graphics drivers – Windows/Linux Baremetal, Pass-through and vGPU drivers||√||√|
|GRID vGPU host software||√||√||√|
|VMware vSGA driver||√|
gpumodeswitch tool for
|NVIDIA License Manager||√||√||√|
|1 GRID Virtual App
Edition for RDSH App
|NVIDIA GRID Certified Servers|
|GPU Card||Tesla M10||Tesla M60||Tesla M6|
|NVIDIA Certified Server Compatibility URL||Tesla M10||Tesla M60||Tesla M6|
nVIDIA GRID Profiles & Configuration Matrix
API, Open-source programming Language Support Matrix for GRID with VMware, Citrix and Microsoft
*GPU pass-through requires special hypervisor enablement which is available in VMware vSphere Hypervisor (ESXi) and Citrix XenServer.
*Windows Server 2016 supports NVIDIA GPU pass-through with Discrete Device Assignment (DDA).
nVIDIA GRID licensing, Packaging and User Guides
- NVIDIA GRID Packaging and Licensing Guide
- GRID Licensing User Guide
- GRID Virtual GPU User Guide
- For latest NVIDIA GRID updates refer the detailed blogs by @
- Where to download nVIDIA GRID software? As per the nVIDIA portal, M6 and M60 require the purchase of NVIDIA GRID software, download your software licenses here
- Where to find the drivers for NVIDIA GRID 2.0 vGPU and up, including M60 and M6 drivers
VMware GPU References
- Graphics Acceleration in View Virtual Desktops
- VMware Horizon 6 3D Engineering Workloads Reference Architecture
(Note: The reference architecture refers to K1/K2 cards, therefore considering the latest updates and changes, refer to the latest release notes of Horizon View 7.x and nVIDIA GRID M6/10/60 guides)
VMware Terminologies for Virtual GPU solutions
- NVIDIA GRID vGPU (shared GPU hardware acceleration) Available with vSphere 6.0 and later, this feature allows a physical GPU on an ESXi host to be shared among virtual machines. This feature offers flexible hardware-accelerated 3D profiles ranging from lightweight 3D task workers to high-end workstation graphics power users.
- Virtual Dedicated Graphics Acceleration (vDGA) Available with vSphere 5.5 and later, this feature dedicates a single physical GPU on an ESXi host to a single virtual machine. Use this feature if you require high-end, hardware-accelerated workstation graphics.
- Virtual Shared Graphics Acceleration (vSGA) Available with vSphere 5.1 and later, this feature allows multiple virtual machines to share the physical GPUs on ESXi hosts. This feature is suitable for mid-range 3D design, modeling, and multimedia applications.
- Soft 3D Software-accelerated graphics, available with vSphere 5.0 and later, allows you to run DirectX 9 and OpenGL 2.1 applications without requiring a physical GPU. Use this feature for less demanding 3D applications such as Windows Aero themes, Microsoft Office 2010, and Google Earth.
VMware VMotion Capability: Because NVIDIA GRID vGPU and vDGA use PCI pass-through on the ESXi host, live VMotion is not supported. vSGA and Soft 3D support live VMotion.
Microsoft GPU References
- RemoteFX vGPU Updates in Windows Server Next
- RemoteFX vs DDA in Windows Server 2016
- How to Install latest version of DirectX and hotfixes on windows client
Note: Based on my search online, there are not proper official documentation by Microsoft on RemoteFX, GRID or otherwise. And the blogs are quite old and relies mostly on results gathered from various customer environment and user community.
Citrix GPU References
- GPU acceleration for Windows Desktop OS
- GPU acceleration for Windows Server OS
- Configuring XenServer 7.0 for Graphics
Notes from Citrix docs:
Support for NVIDIA Kepler architecture. HDX 3D Pro supports NVIDIA GRID cards (see NVIDIA GRID) for GPU pass-through and GPU sharing.
NVIDIA GRID vGPU enables multiple VMs to have simultaneous, direct access to a single physical GPU, using the same NVIDIA graphics drivers that are deployed on non-virtualized operating systems.
Support for VMware vSphere and VMware ESX using Virtual Direct Graphics Acceleration (vDGA) – You can use HDX 3D Pro with vDGA for both RDS and VDI workloads.