GPU acceleration for standard applications?

We have a XenApp farm with mainly standard programs like Office and Adode Reader, but also some WPF-based in-house developments to view statistics and graphs.
Now that we’re about to buy new servers, the question of GPU acceleration came on.

Will GPU acceleration be of any use for such applications?
Will the user see a difference, or is it reasonable only for CAD and 3D applications?

Everything seems to be moving towards better visuals, and why would we expect anything different? Office 2013, as an example, uses GPU Acceleration by default. Look to the latest PowerPoint animations for proof of the increased use of GPU. WPF supports the use of GPU, Direct3d and such, so it depends on how its built I assume. My favorite recent example of common application using GPU that gets overlooked is Chrome delivered Google Earth. It shows up as Chrome on application inventories and thus gets dismissed, end users then complain. We are working on metrics around common apps, to augment the work we are doing around graphics intensive apps, because its clear GPUs are making a difference across the board. A trial/POC is the best way to know for sure.

Google Earth is a good example, because you can also pick if you want to use openGL or DirectX acceleration. I think we even see some improvements in video streams, like Flash video (youtube).
Other apps like ArcGIS also profit from graphics acceleration, in addition to of course the obvious CAD/CAM apps. The other thing to bear in mind is that the CPU load typically drops between 30 - 50% in our tests, depending on the application, so you will get more performance or potential number of VMs from your hosting server.

To tjkreidl’s point, we have subjective reports of Flash and streaming video improvement (I personally have demonstrated it for VDI customers and had it work perfectly, to their great happiness to finally have an answer), but we still have no metrics to be objective about it. tjkreidl also makes the CPU savings point, and you can adjust your density math as you scale up your deployment based on the savings you actually see. All good things, now we need to find metrics to satisfy the stats folks.