Hi Jooji,
2000 concurrent users is easy to fit into 2 racks, you can probably squeeze it in one, but it'll be full. I'd suggest against blades (my preference) but if that's something you're comfortable with, all the power to you. I would suggest against a hyperconverged solution for VDI.
In Win10 VDI - you absolutely want GPU everywhere. Win10 runs boatloads better with a hardware graphics processor. But keep in mind - you need fast CPU's if you're going to use GPU. CPU becomes the bottleneck for VDI VM performance when you add GPU.
If you have any questions, please ask away.
VDI Engineer
VCP-DCV, VCP7-DTM, VCAP7-DTM Design