Martin Garton wrote:
> > Choice 3), which I have suggested before, is to choose a
> > widely-supported (and efficiently emulatable) video device and emulate
> > that.
>
> Is that ever going to be as efficient as emulating some theoretical
> hardware that is optimised for virtualisation?
Perhaps not, in theory. To really answer that you would have to look in
detail at the particular achitecures in question. However, I suspect it
could be very close. Certainly it would be close enough for a first
step.
In practice it might turn out to be more efficient. See below.
> The latter seems to be the (somewhat succesful)
> approach that has been taken by vmware.
Actually, in practice I am not very impressed with their approach. From
what I've seen, their Windows drivers just aren't that good (and they
don't support DirectX, at all).
It is a *lot* of work to do good Windows drivers, and this is a really
good reason for using off-the shelf drivers if possible. They're better
optimized and more complete. The end result would likely be
considerably more efficient than lame drivers written for some
plex86-only video device.
