Artificial Intelligence Computing Leadership from NVIDIA
RDP client + NVDIA Tesla M60 Grid card
Hello everyone, I'm planning an environment to delivery Win10 VDIs that will need to use 3D apps like AutoCAD and image processing. The customer would like to use Win10 VMs and Citrix Hypervisor to host them. In the user layer, they will receive a thinclient and will use a RDP connection to connect to Windows 10 VM (virtualized in Citrix Hypervisor with Tesla Grid configured). I've saw this in the NVidia documentation: [color="gray"][justify] [b]Using Windows Remote Desktop (RDP) to access Windows 7 or Windows Server 2008 VMs running NVIDIA vGPU will cause the NVIDIA driver in the VM to be unloaded[/b]. GPU-accelerated DirectX, OpenGL, and the NVIDIA control panel will be unavailable whenever RDP is active. Installing a VNC server in the VM will allow for basic, low-performance remote access while leaving the NVIDIA driver loaded and vGPU active, but for high performance remote accesses, use an accelerated stack such as Citrix Virtual Apps and Desktops. From <https://docs.nvidia.com/grid/latest/grid-vgpu-user-guide/index.html#performance-monitoring-gpu-nvidia-smi> [/justify] [/color] Customer must use Citrix Virtual Apps and Desktops to delivery the VDIs?
Hello everyone,
I'm planning an environment to delivery Win10 VDIs that will need to use 3D apps like AutoCAD and image processing. The customer would like to use Win10 VMs and Citrix Hypervisor to host them.

In the user layer, they will receive a thinclient and will use a RDP connection to connect to Windows 10 VM (virtualized in Citrix Hypervisor with Tesla Grid configured).

I've saw this in the NVidia documentation:

Using Windows Remote Desktop (RDP) to access Windows 7 or Windows Server 2008 VMs running NVIDIA vGPU will cause the NVIDIA driver in the VM to be unloaded. GPU-accelerated DirectX, OpenGL, and the NVIDIA control panel will be unavailable whenever RDP is active. Installing a VNC server in the VM will allow for basic, low-performance remote access while leaving the NVIDIA driver loaded and vGPU active, but for high performance remote accesses, use an accelerated stack such as Citrix Virtual Apps and Desktops.

From <https://docs.nvidia.com/grid/latest/grid-vgpu-user-guide/index.html#performance-monitoring-gpu-nvidia-smi>




Customer must use Citrix Virtual Apps and Desktops to delivery the VDIs?

#1
Posted 01/02/2020 05:27 PM   
Hi If you're using Citrix, then you won't be using RDP (RDP is a Microsoft Protocol), you'll be using "HDX 3D Pro" (which is a Citrix Protocol) and there's no need to use VNC either. Also, don't use M60 (Maxwell) GPUs in a new environment. Anyone building a [b]new[/b] environment should be looking at T4 or RTX 6000 / 8000 GPUs, or a mix of them depending on the environment. These will be more cost effective, more flexible and provide much better performance. Although they will technically work in a new environment, Maxwell and Pascal architectures should now be used for [b]existing[/b] environments where those specific architectures / GPUs have already been deployed and consistency is required for scaling up / out (although there's an argument against doing that as well), and Volta is altogether a different use case. It sounds like you're quite new to this type of technology. You should be planning to do this already, but I would strongly advise you run a POC with the customer and also get yourself a private test Lab so you can experience and familiarise yourself with these technologies before discussing them with customers first, it will make things a lot easier for you :-) Regards MG
Hi

If you're using Citrix, then you won't be using RDP (RDP is a Microsoft Protocol), you'll be using "HDX 3D Pro" (which is a Citrix Protocol) and there's no need to use VNC either.

Also, don't use M60 (Maxwell) GPUs in a new environment. Anyone building a new environment should be looking at T4 or RTX 6000 / 8000 GPUs, or a mix of them depending on the environment. These will be more cost effective, more flexible and provide much better performance. Although they will technically work in a new environment, Maxwell and Pascal architectures should now be used for existing environments where those specific architectures / GPUs have already been deployed and consistency is required for scaling up / out (although there's an argument against doing that as well), and Volta is altogether a different use case.

It sounds like you're quite new to this type of technology. You should be planning to do this already, but I would strongly advise you run a POC with the customer and also get yourself a private test Lab so you can experience and familiarise yourself with these technologies before discussing them with customers first, it will make things a lot easier for you :-)

Regards

MG

#2
Posted 01/03/2020 09:48 AM   
Thanks for your reply MrGRID! so, just to clarify.. costumer doesn't have Citrix Virtual Apps at this moment, so, I'll propose a new lab with this technology and nvidia to them. Thank you so much to advance.
Thanks for your reply MrGRID!

so, just to clarify.. costumer doesn't have Citrix Virtual Apps at this moment, so, I'll propose a new lab with this technology and nvidia to them.

Thank you so much to advance.

#3
Posted 01/03/2020 03:58 PM   
Hi Before you design the POC environment, you need to gather some details and metrics from their existing physical workstations. You should have the specifications of the physical workstations (CPU, RAM, GPU, Storage), the resolution of the monitors and how many monitors are going to be used, a full list of all applications, and also some newly captured performance and utilisation metrics from these applications. That’s an absolute minimum. After you have those details, you can begin configuring the appropriate hardware specifications for the POC. At that point, you can also find out whether they are experiencing any performance limitations, you can then add performance where it’s needed into the POC specifications. As you’ve already mentioned AutoCAD and Image Processing, your CPU choice should have a fast base clock (3.0GHz +). Forget about Turbo, you’ll have a better, more consistent experience with a fast base clock. Depending on the application requirements (how it scales it’s performance) and how many users you plan to have on each Server, you may want to use a pair of these: https://ark.intel.com/content/www/us/en/ark/products/192451/intel-xeon-gold-6254-processor-24-75m-cache-3-10-ghz.html A 12 Core variant of the 3.1GHz CPU is available. Also, there’s a 3.3GHz 12 Core variant as well. Which version you go for will depend on how the applications perform, but your minimum base clock should be 3.0GHz. For the Image Processing, pay close attention to which components are used (whether it’s CPU (single threaded or multi threaded (if multi threaded, how many Cores can it make use of))) or does it use the GPU. If it uses the GPU, how much framebuffer and processing are used. If it’s heavy, you may want to look at an RTX 6000 / 8000 over a T4. As a heads up, currently, the latest vGPU release (vGPU 10) doesn’t support the latest XenServer release (XenServer 8.1). So when you’re ready to build the POC, make sure you check the current supported versions and if needed install XenServer 8.0. That will allow you to run XenDesktop 1912, Windows 10 1909, and vGPU 10. You can then simply upgrade XenServer to 8.1 once it’s supported. Regards MG
Hi

Before you design the POC environment, you need to gather some details and metrics from their existing physical workstations. You should have the specifications of the physical workstations (CPU, RAM, GPU, Storage), the resolution of the monitors and how many monitors are going to be used, a full list of all applications, and also some newly captured performance and utilisation metrics from these applications. That’s an absolute minimum. After you have those details, you can begin configuring the appropriate hardware specifications for the POC. At that point, you can also find out whether they are experiencing any performance limitations, you can then add performance where it’s needed into the POC specifications.

As you’ve already mentioned AutoCAD and Image Processing, your CPU choice should have a fast base clock (3.0GHz +). Forget about Turbo, you’ll have a better, more consistent experience with a fast base clock. Depending on the application requirements (how it scales it’s performance) and how many users you plan to have on each Server, you may want to use a pair of these:

https://ark.intel.com/content/www/us/en/ark/products/192451/intel-xeon-gold-6254-processor-24-75m-cache-3-10-ghz.html

A 12 Core variant of the 3.1GHz CPU is available. Also, there’s a 3.3GHz 12 Core variant as well. Which version you go for will depend on how the applications perform, but your minimum base clock should be 3.0GHz.

For the Image Processing, pay close attention to which components are used (whether it’s CPU (single threaded or multi threaded (if multi threaded, how many Cores can it make use of))) or does it use the GPU. If it uses the GPU, how much framebuffer and processing are used. If it’s heavy, you may want to look at an RTX 6000 / 8000 over a T4.

As a heads up, currently, the latest vGPU release (vGPU 10) doesn’t support the latest XenServer release (XenServer 8.1). So when you’re ready to build the POC, make sure you check the current supported versions and if needed install XenServer 8.0. That will allow you to run XenDesktop 1912, Windows 10 1909, and vGPU 10. You can then simply upgrade XenServer to 8.1 once it’s supported.

Regards

MG

#4
Posted 01/04/2020 12:50 AM   
Hi Just to add ... NVIDIA have now added support for XenServer 8.1 and vGPU 10, so you can ignore the last paragraph above. Regards MG
Hi

Just to add ... NVIDIA have now added support for XenServer 8.1 and vGPU 10, so you can ignore the last paragraph above.

Regards

MG

#5
Posted 01/09/2020 08:45 AM   
Scroll To Top

Add Reply