Maximum PC - USA (2022-06)

(Maropa) #1
00
AVERAGE FRAMES PER SECOND OVER THE GAMES TESTED

20 40 60 80

Back in 2015, when the newly released
R9 390 was running the games of that era,
there was no question about which was
faster—the Nvidia card was consistently
about 10 percent faster. Fast forward to
today and the R9 390 with its 8GB now
leads the GTX 980 by 3 percent at 1080p,
and the lead grows at 1440p (not that
either card is really playable at 1440p
ultra these days). Similarly, the Fury X
used to be nearly 30 percent faster than
the 390, but now that lead has shrunk to
19 percent, and the two cards are virtually
tied for performance in Far Cry 6 and
Forza Horizon 5.
Cards with 6GB of memory strike an
interesting middle ground. There are
still times when it’s not quite sufficient
for ultra settings, but the GTX 1060 6GB
as an example outpaces the GTX 980 by
12 percent in our modern suite of games,
where it was a few percent slower than its
predecessor back in 2016.
Going forward, we’ve mostly reached
the end of the road for 4GB cards. They
still have a place in the budget sector,
but the RX 6500 XT would have benefited
greatly from having more memory and a
wider memory interface. 8GB cards are

AMD AND


NVIDIA GPU


FEATURES
The latest generation GPUs from AMD
and Nvidia both support ray tracing and
the full DirectX 12 Ultimate feature set,
which defines the baseline of what we
expect to see going forward. Nvidia still
has some extra hardware in its tensor
cores, which power technologies like
DLSS (Deep Learning Super Sampling)
and Nvidia Broadcast, and those are both
useful additions.
AMD offers something of an
alternative to DLSS with its FSR
technology, which is due for a major
update in the not-too-distant future. FSR
2.0 will move to a temporal upscaling
solution, but it will still work on a wide
range of GPUs, including those from the
competition.
Intel meanwhile has XeSS (Xe Super
Sampling), which sounds an awful lot like
DLSS, only done on Intel hardware. But
having a software solution such as DLSS
or FSR isn’t the same as getting it adopted
by game developers, and DLSS easily
wins in that arena.
Stepping back to earlier generation
GPUs can mean giving up features, and in
some cases, certain games may refuse
to run at all. Of our test suite, Far Cry 6
was the only game that didn’t want to run
on one of our legacy GPUs. The GTX 780
lacks the appropriate DirectX feature
level, and the game simply failed to run
despite our best attempts. Of course,
there are also games such as Metro
Exodus Enhanced Edition that require ray
tracing hardware, so you’ll need at least
an AMD RX 6000-series or an Nvidia RTX
card to run it.
Other features that ultimately deliver
more performance, like AMD’s Infinity
Cache, aren’t particularly important to
the end-user, as speed is often the goal.
However, video codec support can be
important, especially if you’re thinking of
using a card in a home theater setup.
Nvidia currently delivers better
overall video encoding and decoding
hardware, though Intel’s upcoming Arc
GPUs will be the first with hardware
AV1 encoding and decoding support.
Nvidia’s Ampere and Turing GPUs support
hardware HEVC and AVC encoding and
decoding, except for the GTX 1650.
AMD’s 6000- and 5000-series GPUs
also support hardware HEVC and AVC
encoding and decoding, except for the
RX 6500 XT.

It doesn’t make sense to test
most graphics cards at 4K and
ultra quality, as the vast majority
of older GPUs will gurgle and die
while trying to choke down such
a workload. However, for those
that have the requisite display
and want to know what sort of
graphics card they should use,
we tested a subset of the fastest
18 GPUs from the past five years.
The RTX 3090 Ti naturally
claimed top honors, with an
even greater lead over the
competition than at lower
resolutions—CPU bottlenecks
are no longer a factor at 4K. It’s
10 percent faster than the RTX
3090, and 20 percent faster than
AMD’s RX 6900 XT.
More importantly, the
previous generation’s top GPUs
came in well below the 60fps
mark, with 48fps on the Titan
RTX, 45fps for the RTX 2080 Ti,
and 39fps from the RX 5700 XT.
Going back one more generation,
the Radeon VII and GTX 1080
Ti barely managed a 30fps
average, with multiple games
falling well below that mark.

still generally safe, but there are certainly
games where you’ll want to drop a few
settings (like texture resolution) to avoid
memory thrashing—where there’s too
much data needed in the GPU’s VRAM,
and it ends up having to pull in data over
the much slower PCIe bus. At the same
time, going beyond 12GB definitely isn’t
needed in our current test suite, and
while it’s inevitable we’ll eventually see
consumer graphics cards with 32GB or
more memory, that’s many years away
from becoming close to mainstream.

FOUR YEARS OF RAY TRACING
Who doesn’t love eye candy? With its
GeForce RTX graphics card launch in
2018, Nvidia ushered in the era of real-
time ray tracing for games. Since then, the
rate of adoption has been more of a slow
burn. We all know what ray tracing—or
the more advanced path tracing—can do
for movies, but getting similar results at
real-time framerates remains difficult.
We’ve selected the best showcases
for ray tracing in games, with multiple
RT effects including shadows, global
illumination, reflections, ambient
occlusion, and caustics. Ironically,

MAXED OUT AT 4K


RX Vega 64

RTX 5700 XT

GTX 1080 Ti

Radeon VII

RTX 2080 Super

RX 6700 XT

RTX 3070

RTX 2080 Ti

RTX 3070 Ti

Titan RTX

RX 6800

RX 6800 XT

RTX 3080

RX 6900 XT

RTX 3080 12GB

RTX 3080 Ti

RTX 3090

RTX 3090 Ti

4K ULTRA GPU HIERARCHY

23

29

30

31

35

39

43

45

47

48

51

59

61

63

66

66

69

76

JUN 2022 MAXIMU MPC 33

Free download pdf