A few weeks ago I wrote a KB article outlining the support on various virtualization stacks for hardware encode support.using NVIDIA NVENC functionality. I’m now in the process of having the KB article updated as today Citrix announced with XenDesktop/XenApp 7.11 that they have added support for Windows alongside their existing Linux support.
A few vendors already offered hardware encode for both Linux and Windows such as NICE DVC and HP RGS, however prior to this release from Citrix:
- Citrix only supported Linux
- VMware only supported Windows (and then only single monitor)
This sees Citrix now offer multi-monitor support on both Linux and Windows. I expect VMware will now be under some pressure to match.
Many customers use both Windows and Linux so this parity makes life much easier for the system administrator.
VDI can use GPUs in many ways:
- To accelerate the applications
- For the protocol encoding and decoding e.g H.264 e.g. NVENC
- Allowing protocols access to the frame buffer directly
So solutions without NVENC also do still benefit from GPU availability.
You can read about the HDX/XenDesktop 7.11 enhancements here: https://docs.citrix.com/en-us/xenapp-and-xendesktop/7-11/hdx/gpu-acceleration-desktop.html
There are also some other big leaps in HDX such as the addition of hardware H.264 4:4:4 encode (removes the artefacts of standard H.264 4:2:0 but at a cost of a bit more bandwidth).
Nicely any Citrix Receiver that also supports H.264 decoding can be used with NVENC hardware encoding (H.264 4:2:0). Which should ease users trialing and adopting.
The protocol war rages on!
Multi-monitor NVENC hardware encode for both Linux and Windows is definitely welcome news!
Given their slow release cycle, I wouldn’t expect VMware to be able to match this anytime soon.
This needs to be augmented with support for the same on XenApp connections (even if only for a single monitor)!
Support definitely needs to be added for GU passthrough and XenApp.
Support definitely needs to be added for XenApp and GPU passthrough! It should be possible at least if using an “A” type configuration, which already leveraged a vGPU, e.g. M60A8.