FEAATURE//ANALYSSIS
mesandperforming post-processing.
tead, it’s an enhancement of the classic
AA method. Here, the temporal element
is the fact that each alternate frame uses a
different sampling pattern, and MFAA then
combines the two frames.
So, for conventional MSAA, you may have
four sample points per pixel, which are averaged
out in each frame. However, for MFAA, you only
have two sample points per pixel per frame,
but the points move to a slightly different
position in the next frame. If you average the two
results, you get the effect of 4x MSAA with the
performance hit of 2x MSAA.At least that’s the theory, but we found
the difference in image quality between
TAA and MFAA to be subtle to the point of
imperceptible. Meanwhile, the performance
difference was minimal, with less than a 2 per
cent improvement in the frame rate for any
given AA sample rate. Like other temporal-
derived AA methods, we also felt like MFAA
made for a slightly less responsive-feeling
experience – if you have the hardware for it,
MSAA is better for first-person shooters and
other faster-paced games.Deep Learning Super
Sampling (DLSS)
The latest innovation in AA comes courtesy
Nvidia’s very latest RTX graphics cards,
and in some ways it’s the direct opposite
of so many conventional AA techniques.
Instead of sampling at a higher rate and
downsampling, DLSS all about upscaling
from a lower resolution.
It works in three steps. First, Nvidia submits
screenshots of a game to a neural network,
taken with up to 64x SSAA applied, along with
screenshots with no AA applied. It then gets
the network to learn how to best approximate
the 64x SSAA images from the 0x AA
images. This trained data is then packaged
into Nvidia’s drivers, where it’s used by the
Tensor cores of RTX cards to rapidly apply the
best techniques to upscale any given game.
When the game runs with DLSS enabled,
the system dynamically reduces the
resolution of the scene then upscales it using
the learned DLSS optimisations. In this way.
you can run a game at an apparent native
4K resolution, say, but with up to twice the
frame rate. The results are impressive. with
image quality that rivals TAA but without
TAA’s problems with fast motion. However,
there are two big caveats. The first is that it’s
only available for games that have built-in
support, with Nvidia also having to run the
optimisations on its side and upload them to
its driver. Currently, there’s just a couple of
dozen games that support it.
The other caveat is that it’s only really useful
on games that don’t require a fast frame rate.
Because the algorithm uses the Tensor cores,
it’s intrinsically linked to how fast they can
work. As such, DLSS is best used for getting
maximum image quality in slower-paced
games, rather than getting slightly better
image quality in faster-paced games.Top: Final Fantasy XV at
4K with TAA
Bottom: Final Fantasy XV
at ‘4K’ with DLSS