[mythtv-users] Is NVIDIA worth the bother?

[David Brodbeck]

On Thu, October 1, 2009 10:03 am, Eric Sharkey wrote:
> The idea is to isolate the "code that writes directly to the memory
> and registers of the video card" from the "code that figures out what
> to write".  Presumably the former code can be open source and the
> latter code closed.  If there's a secure message passing interface
> between the two, this can be done safely, although there may be some
> performance hit with this kind of architecture.

I seem to recall that Windows NT tried this -- putting the video driver
outside of the kernel -- and reversed it in later versions because the
performance hit was so bad.  Video drivers have to move a lot of data. 
(Anyway, if the video driver crashes but leaves the kernel up there's not
much you can do with a Windows system except hit the big red switch.)

On a higher level, this is still the main model for X Windows -- the
client program talks to the X server via a socket protocol and the server
does all the driver-level interaction.  This turns out to be way too
inefficient for things like video, which is why we now have such a variety
of direct-rendering and shared-memory interfaces for X.