Originally Posted by netllama
The NVIDIA driver only supports one X server per GPU.
This is truely bad news!
From the very early nvidia driver versions (mid of 2000) up to 1.0-7xxx,
i had realy bad experiences concerning switching VTs or runnig multiple
X server instances. This happend with various Linux distributions and
combinations of Intel/AMD and nvidia hardware, which made me assume
that there is a subtle software bug. I was never able to track down this
extremly annoying problem. Every time i needed to switch to another
VT, i had to pray that the box doesn't get stuck! However, this issue
magically disapeared for me on my current HW setup since some early
1.0-8xxx driver version - I didn't had any crashes for more than a year.
It sad to hear that there is regression for Fedora 7!
From what i read in this forum, this problem doesn't show up with the
"nv_drv" driver that comes with the X server. Since the crashes with
"nvidia_drv" do also happen when no 3D is active at all, i assume the
problem is somewhere in the "2D" part where the GPU context is saved
for switching between VTs. How about re-designing the driver so that
OpenGL hooks in as module for the nv driver? This way, such kind of issue
could be tracked down much more easily. In my wet dreams, i see an
open source nv driver that i.e. uses CUDA for 2D/3D acceleration where
DMA and IRQs are fully handled in user space ...