By Fredrik Brattstig @virtualbrat
21 December -21
Look what I got in the mail!
My friends at NVIDIA recently sent me a NVIDIA RTX A6000 GPU for some testing so lets dig in!
The new NVIDIA Ampere architecture are now implemented in a whole armada of GPU’s, ranging from consumer graphics, like the NVIDIA GeForce RTX™ 30- Series, to the workstation boards (where the A6000 is the flagship), to Server Boards for VDI, compute and AI.
I mounted the A6000 card in one of my LAB servers, running on Citrix Hypervisor 8.2. I installed the NVIDIA vGPU 13.1 (Driver version in VDA is 470.82). I assigned a NVIDIA RTXA6000-8Q profile to the machine and powered it on, no game. The VM didnt power on. Going back to the drawing board, or at least reading the documentation, I figured out that I needed to change the mode of the board, but what does that mean?
The A6000 is a workstation board, so the general design of that board is that it should be mounted in a workstation machine, having up to four high-end displays connected to its DisplayPort outlets. With this new architecture of the boards, when I want to virtualize the GPU power and make it a vGPU beast, the display mode in the microcode of the graphics card needs to be changed to display-less mode.
The NVIDIA Display Mode Selector Tool can be requested at https://developer.nvidia.com/displaymodeselector
When I got that tool, I copied it over to the Citrix Hypervisor host, and ran the tool, it complained that the GPU was busy, so I ended up un-installing the vGPU drivers, ran the tool successfully to update the microcode of the GPU, rebooted the Citrix Hypervisor, installed the NVIDIA vGPU 13.1 drivers again – GAME! It was all working now!
Before we dig deeper, here are the characteristics of the A6000 card
As previously stated, I assigned a NVIDIA RTXA6000-8Q profile which means that I’m splitting this card in eight slices. 8Q means that i have 8 Gigabytes of graphics memory assigned to my virtual machine. The graphics-card has 48GB frame-buffer as you can see in the table above, which means I can have six machines sharing the same card (48/8=6). The VM will be using a NVIDIA RTX Virtual Workstation license to function in optimal state.
As I’ve read articles from a friend about the progress made with Citrix Virtual Apps and Desktops (CVAD) 2112, I decided to set that as a default for my backed systems, and I did choose to use the Citrix Workspace App for Linux (CWAL) 2112 to make sure all was in sync.
The awesome post from Muhammad Dawood@Citrix about the CVAD 2112 3D Pro evolution that can be read here:
https://www.citrix.com/blogs/2021/12/16/hitting-new-highs-with-hdx-3d-pro/
Lets have a look at what I got as results using IGEL OS with CVAD 2112 and NVIDIA A6000 vGPU!
By using optimal network connections, I recorded a few videos. This is all running on my LAB network and have high speed connectivity between my IGEL OS endpoint and the servers. Later on in this blog I’ll do a less optimal connection to see the results.
I have been using Unigine Superposition for graphics benchmark, and by adding the Citrix real-time statistics to the IGEL OS endpoint, I get some interesting session information, reported by the endpoint.
So, lets have a look at the first video. I have a IGEL OS endpoint connected to a 4K screen (3840×2160 pixels) and running the real-time rendering benchmark:
I can confirm that with a performant back-end setup CVAD can push 120 FPS to a
IGEL OS endpoint running CWAL 2112. That is some quite amazing performance. The A6000 and CVAD 2112 delivers a constant 120 FPS in 4K resolution, 4K equals to four FullHD monitors – Speaking of that, lets have a look at another video, where I play with CVAD feature Virtual Displays which allows me to split my 4K screen in four separate smaller screens, yea, FullHD each, four of them!
As seen in the video, THREE simultaneously rendered and displayed video stream’s using different technologies – real time raytrace rendering + realtime rendering + youtube playback – each running in its FullHD 1920×1080 virtual monitor. Awesome power right!? The synergy of NVIDIA A6000 + CVAD 2112 + IGEL OS with CWAL 2112!
Oh, yea, sure, that is optimal network and yada yada.. Do you want to do this remote? Covid seems to be rolling in, again, and employees are starting to be sent home again, “- my designers need to to this remote”.
Lets have a look at less optimal network connections? In the below video, I’m using the same IGEL OS endpoint, disconnecting the LAN cable, and connecting to my iphone hotspot. connected to a 4G cellular connection -> to my netscaler ADC -> CVAD host. Lets have a look
As you could see in the video, there are some slight reductions of frames, and some slight degradation to color depth, but over all the session was very fluent and there was no problems at all to interact with the session. When the complete screen is updated all the time, like when playing a movie, or playing real time rendered content, the amount of bandwidth used increases and that is why we see a slight degraded user experience. It is more than good enough to sit a complete workday with very high demands of a well performing remoting solution.
That’s it for today’s blog post, though, as the A6000 is a nice toy i will spin it a few more rounds, so stay tuned for more blogs!
By that i would like to wish you all the best holiday season ever! Stay well!
/Fred
VirtualBrat 