I have a customer which are testing 3D VDIs from Azure/Citrix Cloud. The virtual machines are NV6 models with Tesla M60 GPUs, 6 vCPUs and 56GB RAM, and are running Windows 10 with VDA 7.18 and AutoDesk products (AutoCAD, DWG TrueView, Inventor, Vault). The latest WorkSpace App (Receiver) is used from the client side.
I have created a Citrix Policy for HDX/graphics with some basic settings (se image_citrix_policy.jpg).
The problem we’re facing is that text in general is very "blurry" on screen (se image_add_remove.jpg, which shows Add/Remove Programs which is opened from within the Win10 desktop).
When I run the latest version of Remote Display Analyzer (3.0) from the Citrix session, it states that Text Optimization is Disabled (see image_rd_analyzer.jpg). Does anyone know how to enable this feature, or how to remedy the problem with "blurry" text, so that the overall graphics quality is as good as possible, both text and 3D/CAD/multimedia.
you are running H264YUV420 so this is expected behavior. You can change to visally lossless (YUV444) or use Active changing regions policy to mitigate this issue.
Make sure you are running H.265 if your endpoints are capable. You’ll get better performance and lower bandwidth usage across the board.
You can use lossless or actively changing regions to help with compression anomalies and sharpness, but they come with their own trade offs. Lossless will cause more dropped frames by nature, and the selective h.264/265 policies tend to have higher average CPU usage in session while running 3D or video because of the algorithm running to determine which parts of the screen require higher framerates.
We had our users play around and choose which policy they liked, and most went with h.265 full screen compression. They also noticed that using a solid desktop background color seems to reduce the prominence of artifacting in the session. YMMV
unfortunately I don’t agree. You won’t see much improvement with H265 as this is also YUV420 so you still have chroma subsampling and "blurry" text. You are right in terms of bandwidth and slightly increased image quality but as I said only YUV444 or bitmap remoting (Thinwire+ or active changing regions) will help here.
In most of my measurements H265 is even worse compared to H264 in terms of color accuracy. And we do real comparison with multi layer SSIM.
I can’t comment on color accuracy. I have no reason to doubt your findings. From a user experience perspective, h.265 has shown itself to be much more performant, particularly over WAN links.
If I was trying to push picture/text quality at all costs, selective h.264/265 with actively changing regions policies work well, but they consume more bandwidth and drive up the overall CPU usage on your hypervisor due to the increased CPU draw from the CtxGfx.exe. Depending on your user density, this could be not a big deal, or a CPU ready/CPU blocking nightmare.
