I did a VDI PoC for an architecture/engineering firm using AutoCad and all associated GPU intensive apps. Obviously right now, there are essentially two options.
1. Use standard VMs brokered by View and hope that the lack of GPU power doesn't inhibit productivity too much.
2. Use physical workstations each with a PCoIP host card, and let View Broker the connections.
The reality is most of these GPU intensive apps still need real GPU power, so while #1 works pretty darn well, it may falter when the user needs the power. #2 is a very clever way to harnes real GPU power. However, there are enough peculiar behaviors of dedicated PCoIP host cards that this isn't quite up to snuff for the end user. That is what the users of the PoC provided feedback on.
The REAL solution will be via NVIDIA's new VGX adapter against View 5.1 and vSphere 5.1. This will give real GPU power to VMs. Debuted earlier this year, but isn't on store shelves quite yet.
I also posted a 5 part series on testing out VDI on graphic intensive applications. This is for the software company that I work for.
http://vmpete.com/2012/01/05/vdi-for-me-part-1/