Have you checked nvidia-smi to ensure NVENC is being used on the XenDesktop VMs (NVENC helps to reduce latency, so good to know if it’s working or not)? Obviously, now that you’re using XenApp, you need a different set of Citrix Policies because NVENC does not work on XenApp and you’re relying on the CPU to handle the compression.
Mentioned above, have you tried going to Storefront directly yet and not using the Netscalers? Have you tried connecting to a XenDesktop with RDP? What happened? …
In one of your replies you mentioned that AutoCAD is CPU based NOT vGPU based … Everything I have ever dealt with for AutoCAD showed your answer to be true for Revit but not programs such as MEP & IDS. We use MEP, IDS & Revit here and when we first did CITRIX everything worked fine but MEP & IDS … lag was ridiculous. (Had K6000 & GRID K2 cards). When I worked to troubleshoot this I kept running into things that suggested REVIT was working fine because that is mostly CPU based and doesn’t rely heavily on the vGPU like MEP & IDS does.
When applications that are working normally on a local workstation are moved to a virtual environment, accessed remotely and then start misbehaving, as long as the virtual environment is spec’d and designed appropriately (I’m assuming yours is?) and not over utilized, one of the first things you want to do is check whether the protocol is introducing the issue. You can do this by accessing the virtual resource using a different technology (typically RDP will suffice for an initial test just to see if the symptoms persist (although there are others)). Another option is the access mechanism, a lot of folks with Citrix environments use Netscaler, however Netscalers are typically configured once for remote access and then never optimized for the specific workflow. Or a Netscaler VPX will sit on a server with unsuitable and / or insufficient and / or over-contended resources. Changing your access method is another good way to narrow down what is causing the issue.
There can be numerous bottlenecks in the system depending on it’s specification, architecture and utilization, and a bottleneck will typically manifest itself as latency to the end user. You could even go as far as changing the user peripherals for professional grade components, rather than the cheap rubbish that so many companies seem to use (keyboard and mouse (I mentioned those earlier in this post too)) as these are also often over looked.
If you have MEP & IDS on an appropriately designed, spec’d platform, as mentioned, I’d check out the protocol and access methods and see what happens.
So I’ve had a quick play with it, and I don’t appear to have the kind of lag you mention above.
Now, as said previously, I’m not a CAD user, and I know how sensitive these teams are when it comes to latency and interaction, so I’m not going to say it’s perfect. I’ve taken 2 screenshots (attached) both running at 2560 x 1600 resolution.
AutoCAD 2017 Plant 3D - 1.jpg
At this distance, I have just under 1 second of latency before the screen updates after a 1 notch mouse wheel scroll, the further I zoom out, it adds a little more update latency.
AutoCAD 2017 Plant 3D - 2.jpg
At this distance and anything closer, I have no latency after a 1 notch mouse wheel scroll.
I don’t know how to apply Hatching or Orbit (I’ve had a go, but it just tries to analyze everything, I’m obviously doing something wrong), but when I enable wireframe and zoom in and out, I don’t seem to have any issues.
With my limited working knowledge of AutoCAD, that’s the best I can do unless I have step by step instructions. Typically I’d obviously let a CAD user do this, and I’ll just fix what they have issues with. rather than me try to use their application.
Just out of interest, I thought I’d also try it at 4K with wireframe, being completely honest, there is a little bit of compression artifacting. I could probably tune it out, but I just wanted to use the same policies as above. Response times are the same as 2560 x 1600.
