What I'm hoping they'll do eventually is allow for GPU passthrough on desktop clients like VMware Workstation Pro. vSphere (server version of VMware) already does it, as does unRAID & others, but it'd be super-cool to have a second GPU & be able to tie it into a VM. So many scenarios:
1. Grab a cheap spare card for older 98 & XP games
2. Have a workstation Quadro GPU in for business stuff & a dedicated VM for gaming with like a GeForce GTX card
3. Have a family multi-player setup so say the parent could have a dedicated gaming rig & the kiddos could play on a Steam thin client on the living room TV
4. Use the dual CPU-integrated GPU (i.e. Intel or Ryzen) & then a dedicated GPU, so if you want to switch from playing local Steam to virtualized vintage games, you could run the host off the iGPU & the VM off the dGPU. This would be particularly useful for laptops, which already have a built-in software-switching system for the iGPU & dGPU combo setup, so there wouldn't have to be a physical cable swap or anything in that case.
On the server side of things, this has been a bear for me. I started training on the updated stuff for Server 2019 just a week ago. Originally we had say a Terminal Server that would let you remote in. Then they added RemoteFX, so you could get like a multi-core, high-VRAM card like a GIRD card & split the power between users in remote sessions, so you could do video streaming & basic DCC work. Then they added GPU pass-through for dedicated GPU access on individual cards. There's also been PCoIP compression cards like Teradici offered & really great software solutions like HP's RGS system (not a perfect interface, but works awesome functionally!). Currently they're on NVIDIA vGPU software licensing for stuff like NVIDIA Quadro Virtual Data Center Workstation (Quadro vDWS).
And this is where it gets fun...on the latest Windows Server OS (2019), they removed the UI for RemoteFX vGPU setup from Hyper-V Manager, but you can still access it through Powershell, as long as you jump through a few hoops:
1. Make sure you have a RemoteFX-compatible GPU card
2. Install the Remote Desktop Virutalization Host role
3. Enable the Host GPU for RemoteFX vGPU
4. Add the RemoteFX vGPU to your VM
5. Configure the RemoteFX vGPU in the VM settings
I've been buying up used GRID & Tesla cards off eBay to test different scenarios with, as I have a lot of small-shop clients who are on a budget, need to do DCC, and have a lot of end users working at home & could really benefit from a non-insanely-priced & non-overly-complicated system setup. Companies are really pushing for custom server-side accelerated graphics due to COVID. HP took their awesome RGS software (which was originally for HP's only, then they started licensing it out - it's the same idea as a dedicated accelerator card for physical GPU's, just in software form) & made it into a centralized server solution called ZCentral:
Free your power users to tackle compute-and graphics-intensive projects from anywhere with Z by HP Remote Access Solution, a complete hardware and software solution for demanding, remote workflows. Learn more here.
www8.hp.com
So now you can pop some HP workstations (I mostly use the Mini's, which are absolutely amazing devices) into a special rack & have centralized management over accelerated remote access! There are a number of companies doing that for end-users through the cloud now (outside of the usual Amazon cloud servers & whatnot), such as ShadowPC:
Cloud Gaming verwandelt deine Geräte in einen leistungsstarken High-End-PC. Wähle das Setup zum Streamen deiner Videospiele oder Software
shadow.tech
They offer 3 tiers right now: a GTX1080 for $15/mo, RTX2080 for $30/mo, and Titan RTX for $50/mo. Pretty nice that you could spend $180 annually to have access to a GTX1080 gaming rig in the cloud on virtually every device you own, anywhere you feel like dialing in from, without having to spend hundreds or thousands on a dedicated gaming computer! Linus has done a couple projects like that on his channel for local gaming using unRAID & the Lime GPU virtualization system, which is pretty dang neat: