Those who manage mixed environments of traditional XenApp servers and GPU-enabled XenApp servers, are you able to use the same MCS or PVS image with both types of servers? Or do you maintain different images? One with NVIDIA GPU drivers installed and one without?

Can manage a single image, just make sure your build vm / gold image vm has a GPU for driver updates etc.

I noticed the NVIDIA Control Panel auto provisions for all users. I’m assuming I can toast that app if no users need its ability.

Yup, squash it with your squishing tool of choice

I've done it previously, only 1 MCS base image, just enable GPU in VM properties before starting it up, and as James mentioned just have the GPU software installed.

Yeah, what Kindon said. Many, many of those scenarios..

Help me understand Citrix IDD. Instead of using the vendor driver, I can configure Citrix IDD to be the display adapter? Would the vendor drivers needs to be installed if I'm using IDD? Would I ever want to do that?

Citrix IDD will be default and is recommended approach now which we have been doing for a little while now (previously always switched to Nvidia), however Citrix IDD is now required for some things like i think Adaptive refresh now and anything with protected/restricted emails showing correctly (which ironically we are talking to customer about right now with information rights management for oulook emails) . So you still have to install the Nvidia guest drive package for sure if wanting the image at some point for attaching a GPU profile to it. The Display Driver priority order is where the Citrix IDD vs Nvidia comes into play. But that doesn't negate having to have drivers there.

Ok great. That clears that up. I have a project to deploy grid cards and I haven't touched grid for 10 years. The documentation seemed to suggest I don't need the Nvidia software.

@Oz Zy Can you elaborate on the Display Driver Priority? Noted on one set of images I have, NVIDIA GRID shows up correctly in DXDIAG as priority, but on another, Citrix does... Is there a rhyme or reason behind what driver is driving?

@Matthew Mueller there is a reg key that can be created (AdapterMerits) for display driver priority. Over time, guidance and what I do has changed as for a while there I was adding key to have Nvidia higher priority but now with some things like adaptive refresh and protected email capability only with IDD vs Nvidia, we typically leave the default which is Citrix IDD priority. You are still leveraging Nvidia GPU. And yes, you still have to do everything like you would do otherwise - so still the GPU profile set and Display drivers installed, etc. This display priority is just what will be leveraged/prioritized in the Citrix brokered session. Display adapter priority and monitor creation in Citrix sessions

You can't "use both"? Think, IDD for certain applications? I just assume users of applications that fault, you would give the registry for IDD (think Exchange encrypted email consumers), and give the registry for NVIDIA for everyone else?

you aren't really understanding. You aren't turning off GPU acceleration just prioritizing another display driver and still leveraging GPU acceleration

I've noted a performance impact switching between the drivers just now; so you're telling me the IDD driver still leverages NVIDIA GRID?

yes it does

NVIDIA just fingered Citrix for the culprit and confirmed to use IDD to your point https://www.citrix.com/blogs/2024/06/10/introducing-citrix-idd-enhancing-your-hdx-experience/ However comma, I've noted that flipping from GRID driver to IDD, does result in reduced performance and reduced framebuffer usage on the backend which corresponds with the performance regression

This is good information, Matthew. I'll keep this in mind with my GPU testing.

There's gotta be a way to have the SDK acceleration on apps that need it and IDD compatability for apps that need it? Any ideas? Setting registries for users is a large catch-all, but I prefer more granularity, but can't think of a better means to automate

Citrix IDD is a display-only driver meaning the GPU will still be leveraged if available. In all of these instances, the Nvidia GPU will still be leveraged as you would expect assuming the profile is on there and Nvidia guest drivers are installed (and therefore you also see the Nvidia display in the device manager). Citrix IDD of course will work good also without GPU acceleration. If you are doing Citrix IDD without Nvidia drivers and a profile associated then yes, you will not have GPU acceleration and your performance will be back to non-hardware acceleration levels. I have not had any throughput issues with Nvidia GPU acceleration with leaving the display priority with Citrix IDD. There is no situation where just having Citrix IDD display driver that you 'lose' the Nvidia GPU acceleration. I'll let the lead on the graphics side @Rody Kossen here explain the in-depth details further.

I can see the in the PRISM metrics on the Guest VMs IDD does indeed leverage the GPU vs DOD, and, I can see that IDD and NVIDIA drivers invoke the GRID licensing, so your statement is true on the face of it. However comma, I can also tell both in real world testing and synthetic benchmarks, IDD leaves alot of the hardware acceleration "on the table" compared to the NVIDIA driver, so much so that it begs the question am I missing some configuration where IDD's defaults throttle NVIDIA GPU-accelerated VMs? Think I need to adjust some registry key for utilizing provioned GPU framebuffer as we're leverging 2GB profiles which I am seeing a drop in framebuffer use when flipping from NVIDIA to IDD

@Matthew Mueller how are you testing the video memory throughput? So, IDD is a Display Only driver and we fully utilize the NVIDIA GPU. Applications should still select the NVIDIA driver for rendering (as IDD is not a render adapter), thus getting full hardware acceleration. IDD is our preferred and default option as it has benefits like being able to see protected content (email / video etc).

winsat; that's all I have access to as I operate in air-gapped, high security networks in terms of sythentic benchmarks so if it isn't built-into Windows, or RedHat 8? No bueno. I do however have a DEV "island" cluster I can test with in more depth, however, it's pretty evident from even dragging a window, or browsing in Chrome, the NVIDIA driver is superior in real-world performance. Nevermind session re-attach is severly delayed with IDD use vs NVIDIA which we're looking into why on our end as we speak. I feel perhaps there's some element we're missing here, IE misconfig on my end as I trust your statements, but 1+1=3 here in terms of percieved performance and winsat numbers

Which VDA version are you using?

2203 CU4 and CU6, depends which enclave... Was going to ask if newer CVADs IDD drivers were improved

Ah! Yeah, we've made significant improvements to our Graphics stack since 2203 🙂

TTC 3.x supports newer CVAD, but TTC2.x is locked on 2203 max... H264 limits

No, that is not possible

Glad to hear the newer CVADs have improved IDD, we were scratching our heads over here wondering why bother paying for new GPUs, if the driver doesn't use it

I can assure you that we fully utilize the GPU with IDD 🙂

Appreciate it, and again, I didn't doubt, more of questioning where the delta lies, as I just assumed I was missing a piece of the puzzle. That makes logical sense

We have some stuff in the release notes, for example: https://docs.citrix.com/en-us/citrix-virtual-apps-desktops/2311/whats-new#adaptive-refresh-rate

Yup, I see it. Looks like you guys have been busy adding support for acceleration of major OSes, APIs, codecs, which is where I gathered the shortfall was