NVIDIA vGPU - Introduction to NVIDIA Virtual GPU

Editor's Notes

• #3: As more organizations enable digital transformation and Windows 10 they are turning towards VDI. According to analysts, there are 3 factors Contributing to VDI adoption. The first is workplace flexibility and business agility. This means being untethered and being able to work from any device, anywhere. The second is improved security. With the rise in the global workforce coupled with more sophisticated threats, it becomes all the more important that intellectual property and mission-critical data are securely hosted within the data center, and not on end-user devices. And let's not forget these threats can be cyber attacks, but they can also be brought about by nature - hurricanes, earthquakes, etc. And third, VDI also results to lower CAPEX and OPEX from being able to have a cost effective infrastructure that scales out to more users with better manageability. http://www.businesswire.com/news/home/20161111005754/en/Global-Virtual-Desktop-Infrastructure-VDI-Market-Grow
• #4: Application and desktop virtualization solutions have been around for a long time, but their number one point of failure tends to be user experience. The reason is very simple. When applications and desktops were first virtualized, GPUs were not a part of the mix. This meant that all of the capture, encode and rendering that was traditionally done on a GPU in a physical device, was being handled by the CPU in the host. While it was functional for some of the more basic applications it never truly met the native experience and performance levels that most users needed. Then a few years ago, NVIDIA released our virtual GPU solution which allowed you to virtualize a datacenter GPU and share it across multiple VMs. This not only improved performance for existing application and desktop environments, it also opened up a whole new set of use cases that could leverage this technology.
• #5: Windows 10 was developed with hardware acceleration in mind. Hardware acceleration means using a computer’s hardware to do certain tasks and functions faster than is possible using software. Moving graphics rendering from the CPU to the GPU allows better graphics performance. Because Windows 10 is fully accelerated using a GPU by default, or is left to emulate a GPU with software rendering when one is unavailable, the operating system’s full potential can’t be realized without graphics acceleration. Furthermore, for most of the latest the applications, HW acceleration is turned on by default. Apps like Office 365, MS teams , PDF viewers and Web Browsers will generally use HW acceleration for all WebGL, videos, and zoom, pan, scroll operations.   It has become tougher to disable Win10 features during the initial releases. For the later releases, while there are options to disable visual effects and animations to decrease resource utilization, you cannot completely turn off the Metro/Modern UI. Stripping down the system and removing apps and services would result to end user dissatisfaction as free Windows 10 upgrades for consumers drive expectations for the same quality of experience at work to be productive.  
• #6: Beyond Windows 10 and office productivity applications requiring more CPU consumption, nearly everything that the modern digital worker does today has become more CPU intensive. Modern browsers like Chrome and Internet Explorer all enable hardware acceleration by default. Often you need to go under advanced settings if you want to use software rendering. The latest standards like WebGL, currently used in 53% of the top 100 websites, can cause the CPU to hit 100% when just animating a simple scene. PDF viewers are likewise hardware accelerated by default. Video conferencing applications like Skype for Business and streaming sites like YouTube are becoming prevalent across the enterprise and are likewise very taxing to the CPU. Creative and design tools like Adobe Photoshop® have features that simply won’t work without a GPU and features that require GPU for acceleration. And lastly, we are seeing trends in use of multiple, high resolution monitors, which significanltly increase the number pixels required to encode and render, thereby increasing CPU utilization. All of these make GPUs even more relevant today, as GPUs help offload tasks from the CPU to ensure users get a great experience and can stay productive.
• #7: The previous slides showed you how much more graphically intensive Windows 10 has become. It only makes sense that the apps that get upgraded as part of the Windows 10 upgrade also require more GPU resources. We worked our partner Lakeside Software and they have a product called SysTrack. One of the things SysTrack does is allow you to deploy an agent out to desktops that gather data and statistics about what applications are installed, how much memory is there, how much memory is being used – and they allow customers to opt in to share that data anonymously as part of a community program. We looked at Lakeside data for about 3M desktops over 5 years and found that the percentages of applications that require graphics acceleration has doubled in the past five years. The data shown here is comparing common office productivity applications used in Windows 7 to their counterparts used in Windows 10. As you can see, the Windows 10 versions drive a higher level of computer graphics than their counterparts in Windows 7. What is equally more striking is how just within a year from 2017 to 2018, you an see how much more GPU hungry these applications has become. – up to 482% higher for Skype, for example. Professional applications like SOLIDWORKS, CATIA, etc. require GPUs to support graphics acceleration, but what is clear from this data, is that even common business applications need GPU resources to run efficiently. And, of course this will continue to grow.
• #10: Before I go into my presentation, I’d like to spend a minute or two talking about the technology and how different a virtual GPU enabled VDI environment is over a traditional CPU only environment. On the left we’re showing a traditional virtualization environment powered only by a CPU. The virtualization layer (or hypervisor), is installed on the server. The hypervisor abstracts the underlying hardware, making it accessible to multiple virtual desktops or guest VMs running on a virtualized host. When a user is working on a virtual desktop, all of the graphics commands from within the virtual desktop or application are intercepted by a generic driver within the guest OS, and are passed down via the hypervisor, to be captured and rendered on the CPU. This drives up CPU consumption and lowers density On the right, is an NVIDIA virtual GPU enabled VDI environment. Here, we have GPUs in the server, and the NVIDIA vGPU manager software that resides within the hypervisor. The NVIDIA software enables multiple VMs to share a single GPU, provisioning virtual GPUs for every VM. In addition to providing a better user experience, it also enables support for more users on a server because work that was typically done by the CPU, can be offloaded to the GPU.
• #11: Before I go into my presentation, I’d like to spend a minute or two talking about the technology and how different a virtual GPU enabled VDI environment is over a traditional CPU only environment. On the left we’re showing a traditional virtualization environment powered only by a CPU. The virtualization layer (or hypervisor), is installed on the server. The hypervisor abstracts the underlying hardware, making it accessible to multiple virtual desktops or guest VMs running on a virtualized host. When a user is working on a virtual desktop, all of the graphics commands from within the virtual desktop or application are intercepted by a generic driver within the guest OS, and are passed down via the hypervisor, to be captured and rendered on the CPU. This drives up CPU consumption and lowers density On the right, is an NVIDIA virtual GPU enabled VDI environment. Here, we have GPUs in the server, and the NVIDIA vGPU manager software that resides within the hypervisor. The NVIDIA software enables multiple VMs to share a single GPU, provisioning virtual GPUs for every VM. In addition to providing a better user experience, it also enables support for more users on a server because work that was typically done by the CPU, can be offloaded to the GPU.
• #12: Before I go into my presentation, I’d like to spend a minute or two talking about the technology and how different a virtual GPU enabled VDI environment is over a traditional CPU only environment. On the left we’re showing a traditional virtualization environment powered only by a CPU. The virtualization layer (or hypervisor), is installed on the server. The hypervisor abstracts the underlying hardware, making it accessible to multiple virtual desktops or guest VMs running on a virtualized host. When a user is working on a virtual desktop, all of the graphics commands from within the virtual desktop or application are intercepted by a generic driver within the guest OS, and are passed down via the hypervisor, to be captured and rendered on the CPU. This drives up CPU consumption and lowers density On the right, is an NVIDIA virtual GPU enabled VDI environment. Here, we have GPUs in the server, and the NVIDIA vGPU manager software that resides within the hypervisor. The NVIDIA software enables multiple VMs to share a single GPU, provisioning virtual GPUs for every VM. In addition to providing a better user experience, it also enables support for more users on a server because work that was typically done by the CPU, can be offloaded to the GPU.
• #13: Click on each link to launch either the relevant web page or go to the relevant slide To return to this “holding” slide, press the Green BACK Arrows at the bottom of the following slides.
• #14: With our portfolio of virtual GPU technology we enable accelerated productivity across a wide range of users and applications. For the knowledge worker, this would be office applications, browsers, high definition video, etc. For the creative and technical professional, NVIDIA virtual GPU technology enables virtual access to power user applications such as Dassault’s CATIA, Autodesk Revit, Petrel, and others – some of which you can see represented on the right side of this slide.
• #20: Here is what our Tesla (and Quadro) cards look like in real life including the old K1/K2 cards. Taken in the UK NVIDIA lab. Notice that the P4 and T4 are very small single PCI slot cards compared to the others. Also note that the RTX 6000 was originally designed as a workstation card and so has a fan (active cooling) and has video ports out the back. Tesla cards don’t have video ports as they are designed to be used virtualised and connected to via a remoting protocol. Also note that power connections are subtly different from each other due to differing power needs. This is why it’s always best to source a server and card from an OEM. To ensure correct power and cables are used. Refer to each cards “Product Brief” to find exact power cable requirements. Product Briefs: https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/tesla-t4/t4-tensor-core-product-brief.pdf https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/tesla-product-literature/Tesla-P40-Product-Brief.pdf https://images.nvidia.com/content/tesla/pdf/Tesla-V100-PCIe-Product-Brief.pdf https://images.nvidia.com/content/pdf/tesla/Tesla-M10-Product-Brief.pdf https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/solutions/resources/documents1/Tesla-P6-Product-Brief.pdf
• #21: With vComputeServer, the NVIDIA virtual GPU portfolio expands beyond client computing. vComputeServer virtualizes server workloads for AI, Deep Learning, Data Science, and HPC. The NVIDIA V100 and T4 GPUs are recommended for use with vComputeServer, but Quadro RTX 6000, and RTX 8000 GPUs as well as NVIDIA Pascal based GPUs are also supported. In the client computing space, NVIDIA continues to offer the best virtual GPU solution for knowledge workers with GRID vPC/vApps, and creative and technical professionals with Quadro Virtual Data Center Workstation (also known as QvDWS).
• #22: To help understand which vGPU license is right for you, this decision tree guides you through determining what you will need in your virtual environment. If you use CUDA or professional workstation applications for users like creative and technical professionals, then QvDWS is the software license that is right for your use case. If you use a virtual desktop with productivity apps for users like knowledge workers, then GRID vPC is the right software license for you. If you have multiple users sharing a single OS through sessions, then GRID vApps is the license that would fit you. If you are running AI, deep learning, data science or HPC compute workloads, the vComputeServer best fits your workflow. If you are still unsure what is the best vGPU software solution for you, please contact your NVIDIA partner for assistance.