Can someone tell me what this IS for then? What kind of VM use cases, and any idea the relative comparison to a standard desktop gpu? Or would that depend on the number of users?
GPU's do a lot more than just gaming these days; they're essentially just a special type of processor that is good at highly parallel tasks, so using OpenCL you can leverage them for all kinds of complex number crunching for scientific applications such as data processing, physics simulations and so-on, plus things like 3d rendering.
That said, a GPU like this *can* be used for gaming, but it's not intended to compete with putting a high-end R9 card into your desktop. Basically it's intended to move away from each user having a high-end workstation, but instead offloading their compute heavy tasks like rendering etc. onto a central server (or server cluster). This makes a lot of sense since many workstation users aren't fully utilising their hardware 100% of the time anyway.
Say for example that you have a dozen workstations, each with a fairly high-end GPU, a blade server with a few of these will outperform each machine individually for compute performance, but may only be equivalent to say eight machines, but if each of those machines is only utilised about 60% on average, then you still end up with more performance overall. Meanwhile your users don't need a big boxy workstation, and can instead work from small, power sipping NUCs, since you've offloaded all of the real work onto the server.