|
What's New?
Other than the move to DDR rather than DDR2 and the
doubling of the memory bus to 256bits, and of course
the new, quieter cooling, the big feature NVIDIA are
touting for the 5900 Ultra is "UltraShadow"
technology. UltraShadow, despite sounding like it
involves creating more or better shadows, is actually
employed to reduce the number of shadows that require
calculating for any given scene. In fact it's the
kind of "common sense" idea that you can't
believe isn't already being done.
Effectively, Ultrashadow allows a developer to set
a depth bound within any given scene in which shadows
are calculated normally on the basis that they're
important and will be seen. Anything outside this
depth bound is essentially ignored. Now As to whether
or not this works in current games without any extra
code being added I'm guessing not as the white paper
quite categorically states that "UltraShadow
gives programmers the ability to calculate shadows
much more quickly by eliminating unnecessary areas
from consideration." Suggesting there needs
to be definite parameters defined at the programming
stage in order to take advantage of the technology.
On the subject of this kind of labour
saving feature, I wonder if I'm the only one who gets
kept awake at night thinking about ways that graphics
rendering could be improved. Sad I know, but there
are lots of things about current rendering technology
that doesn't seem to make sense to a simple, untrained
street urchin like me. For a start why are we still
using Polygons? Surely for simple geometric shapes
like the cube, sphere, cone, torus, cylinder and so
on it would be considerably more efficient to define
the size and shape as a mathematical function? This
would remove the need for labour intensive tessellation
routines as the size/distance changes. In short, why
can't these simple shapes be dealt with as if they
are single polygons rather than the sum of thousand
of polygons? I realise this isn't possible with current
hardware, and I would guess that at least a couple
of "reference points" would have to be defined
in order to accurately apply the texture and to calculate
lighting values but it must be more efficient in the
long run?
Then there's Anti-aliasing. I was playing
NOLF2 last night when it struck me that even in the
darkest of shadows the GPU was sweating away anti-aliasing
edges that were so low in contrast you could barely
make them out. Why can't we have a smart anti-aliasing
routine that only works on pixels when the neighboring
pixels are sufficiently different in contrast, perhaps
with a slider to set the threshold for that contrast
difference? This would mean that a light edge against
a light background would be ignored while a dark edge
would be anti-aliased fully.
Finally, is it time we saw framerate
capping? Above say 120fps I'd sooner see my card's
power diverted to increasing image quality rather
than increasing the framerate to 200 or 300fps. Using
this scheme you could lower the framerate ceiling
and if needed the detail levels and know that in scenes
that don't tax your GPU and the framerate ceiling
is reached your drivers are dynamically increasing
detail level on the fly. Let's face it, it's when
the action is at its lowest that we tend to be admiring
the scenery anyway so it makes sense. Hey, I haven't
had a rant for quite a while so I'm due one :)
Let's look at the card I'm testing:
|
