Graphic Hardware Review in Horizon View

In the early versions of virtual desktops, VMware faced a brilliant challenge called Desktop Presentation for high- quality users. Servers were not designed to render and deliver high-quality images. It's better to have a quick overview of the technology before continuing. Technology supports high-level graphics in several different phases. The first was to support 3D graphics. In vSphere 5 with View 5.0, there was a rendering of software. This feature allows us to use features such as Aero Windows. But do not forget that it was not strong enough. In the next phase, for a number of users with high-level graphics needs, hardware-based GPU support was provided with vSphere.

Certainly, there was no solution to the problem if the request for a high-performance graphics card in the virtual environment was referred to the IT unit many years ago. But you are living and you can hope that you have a lot of solutions. In the VDI environment, graphics are provided using virtualization solutions. The graphics card driver is part of the Hypervisor. Do not forget, as we use the servers for desktop presentation, we also have the power of their graphics cards to provide virtual machines and desktops, but the GPUs are weak servers and are not designed to provide heavy graphic capabilities, and have a lot of limitations. For use. In today's world, we are seeing a wide range of technological changes every day. With the offer of View 5.2 at the end of 2013, offering hardware graphics turned into one of the features of this product that was presented in two ways:

• Virtual Shared Graphics Acceleration (vSGA)

• Virtual Dedicated Graphics Acceleration (vDGA)

Virtual Shared Graphics Acceleration (vSGA):

Implementing the vSGA capability allows the GPU to be installed on a physical server and shared to a large number of virtual desktops. In this model, the virtual desktops do not have direct access to the server's GPU and, instead, the VMware SVGA 3D Graphic Driver is part of the VMware Tools installed on the virtual desktop operating system. SVGA Driver is a VMware driver that supports Direct X9.0c and OpenGL 2.1 support. The GPU is in hardware on the ESXi server. Instead of installing a graphics card on a virtual machine operating system, this configuration will be placed in the VIB file on the Hypervisor. Graphical tasks (graphic instructions) are downloaded from driver sessions to the Hypervisor. These happen when GPU controls are on the ESXi server.

The diagram below represents the vSGA architecture:

The vSGA supports apps based on OpenGL 2.1 and DirectX 9, and desktops that have Windows 7 or 8.

vSGA supports the following GPU card manufacturers:

• Intel HD Graphics P4700

• Tesla M6 and M60

• Grid K1 and K2

• AMD FirePro S4000X, S7000, S9000, S9050 and W7000

How many virtual desktops are supported by vSGA technology?

One of the questions and concerns of people involved with Horizon is that we can create a desktop with the graphics hardware in a viewing environment? As you know, based on the use of the Horizon View, you can have different pool desks, which we'll explain in the future. When one of the desktop pools is for users who use a lot of graphics cards and have heavy graphics work, you can not afford to access the GPU just for those users. That's why you create a Pool desktop for this particular item.

According to the above description, to answer this question, the number of virtual desktops that you can assign to a GPU is limited to the amount of video RAM (VRAM) that you assign to each desktop. The only thing to keep in mind is the sharing of resources in the virtual environment, the rest is done routinely. Half of the video memory allocated to the virtual machine is from the GPU and the other half is hosted. When you do the initial steps for sizing the server, make sure you have enough memory to the server. Also, the amount of graphics card required to fit the needs should be ordered. According to the above description, and based on the number of virtual desktops and VRAM-based assignments, we need to calculate how many desktops are available.

For example, the default value of the VRAM allocated to the virtual desktop is 128 MB. In this example, 64 MB of GPU and 64 MB of the rest of the server are provided. If you have a 4 GB RAM card, you can have 64 virtual desktops. 4 GB or 4096 MB is divided into 64 MB of GPU, which is equivalent to 64 virtual desktops. In Horizon View, we can eventually assign 512 MB of VRAM to any virtual desktop. If we extend this request to the previous example, the number of machines available can be reduced to 16. 4 GB or 4096 MB is divided into 256 MB of GPU, which is equivalent to 16 virtual desktops. With the AMD brand, the maximum number of supported desktops for each GPU is 15.

Now, think about what happens if your desktops are scarce. If you create Desktop pool more than GPU capacity on server, built-in desktops will not boot ...

Virtual Dedicated Graphics Acceleration (vDGA):

While vSGA works on the basis of sharing. vDGA allows our virtual desktop to be used exclusively from the GPU card installed on the ESXi server. This will make the virtual desktop extremely high in terms of graphics performance. By doing so, you'll see improvements in quality and performance in apps like CAD, and there are DirectX (9, 10, 11), OpenGL4.4 and NVIDIA CUDA support.

The following figure shows the vDGA architecture:

vDGA is a solution using the VMDirect Path I / O Pass-Through feature, known in some cases as PCI pass- through. This feature makes the virtual machines in the hypervisor layer execute the path-through operation and access the hardware directly installed on the server. In this case, the hardware used is the NVIDIA GPU. A virtual desktop connects directly to a GPU (one by one). When you do this, the capabilities of using HA, DRS, and vMotion will be disabled.

How many virtual desktops are supported in vDGA?

Unlike vSGA, which is limited by the amount of memory available on the GPU, vDGA is limited by the number of hardware GPUs or GRID cards available on the server. For example, a NVIDIA GRID K2 GPU has two GPUs on its board, which means we can assign up to 4 desktop units. Depending on the server hardware, you can add more cards. As the number of cards increases, the number of users that can be connected to the virtual desktop is increased.

The following GPUs are compatible with vDGA:

• GRID K2 and K2

• Tesla M60 and M60

• Quadro 1000M, 3000M, and 5000M

• Quadro K2000, K2200, K3100M, K4000, K4200, K5000, K5200, and K6000

• AMD FirePro S7150

Virtual GPU:

In the previous section we explained two different graphic presentation models. In both cases, we saw a series of limitations in both solutions. In vSGA, the number of users who can use the GPU card can be increased. Of course, given the fact that the driver provided is not related to the GPU, it may be difficult to run some programs. But in vDGA, due to the installation of the driver, the GPU builder minimizes the problems, but limits users' costs and costs in this way. A virtual desktop with a dedicated GPU graphics is excellent for the user, but the cost is untenable. In this situation, we are looking for a solution that covers both modes and is cost effective. A solution that has the GPU shared view for scalability and also uses its graphics card drivers.

The following figure shows the vGPU architecture:

Want to download this blog as a PDF document, click the links below:

https://drive.google.com/file/d/1kyMWyck_0l_udV4n040dUAoQdiQ6YxCX/view

https://drive.google.com/file/d/1UnQDcl64ZjKC5neselp3rrYUVdudrWkn/view

Abd El-Rahman Oreiby

Senior Data Center Engineer

Al Thuraya Security Egypt