On Thu, 13 Sep 2001, Dr Andrew C Aitchison wrote:
> On Wed, 12 Sep 2001, Brett Bolen wrote:
>
> > [Xpert] GLX over network...Here is an old email I found. It states that
> > opengl apps do not use hardware acceleration unless direct rendering is
> > used ( with the exception of the closed source nvidia drivers).
> >
> > Is this still the case?
>
> I haven't heard that this has changed, although I am still running 2.2
> kernels, so haven't kept up with the latest DRI.
>
> Essentially, the point is that with DRI, the client thread talks to the
> kernel which talks to the hardware. This bypasses the X protocol,
> and inter-process communication, so is more efficient.
> However that means that there is no need to put code in the Xserver
> thread to talk to the 3D hardware.
The DRI communicates with the X-server, and does so frequently.
Every cliplist change, the last time I looked.
>
> Until somebody gets around to adding that code (possibly by having
> a dedicated DRI process which gets passed the GLX data that comes
> into the server) there is no hardware 3D acceleration for remote clients.
NVIDIA accelerates indirect clients and we don't need a separate
process. The problem with the others is the DRI. The design of
it isn't completely compatible with the idea of having indirect
clients merely be direct clients in the server's process space.
> I believe that hardware accelerated 3D of remote clients was possible
> with the early voodoo cards and XFree86 v3.3. Since these cards were 3D
> only, Mesa could be made to act as a GLX server. I think this might have
> used the Utah-GLX code.
>
> I've been told that Silicon Graphics don't accelerate remote 3D either.
> No doubt this is only true for some of there systems, but it may mean
> that XFree86 isn't alone in this deficiency.
Every SGI system I've seen accelerates remote clients.
Mark.
_______________________________________________
Xpert mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/xpert
