Predicting 1440p Performance of RDNA2
RDNA2 may have a 1440p edge.
First of all, I need to stress that there’s a lot of speculation
here. There’s also a lot of math, but I will be clear what data I’m using and
which assumptions I’m making. To start things off, I’m going to outline a few
caveats.
1)
I’m using Hardware Unboxed benchmarking data
from the 3090 review, here: https://youtu.be/PTs1gHqvcjs
Go watch it. It’s a great review and be sure to subscribe to their channel. I
use their data because they present it in an easy to consume manner, and they
have among the best testing methodology in the business. However, I also
encourage you to use other benchmarks to help round out the picture and get
other opinions. I’m also using these screenshots without permission, but I feel
like they’d probably be fine with the way I’m using them. If they ask me to remove
them I will.
a.
b.
2) I’m making two big assumptions in this article,
and they could both end up being entirely out to lunch. I’m making it clear which
assumptions I’m making so that you can judge for yourself whether I’m onto
something or not.
a.
Big Navi performance. I’ve done more analysis trying
to nail down the performance based on what AMD showed, but there’s far too many
moving parts and unknown variable, and getting a good estimate is a fool’s
errand for reason I will describe. I am making an effectively wild claim based
on everything I’ve seen, and that may be way off.
b.
I’m going to assume that RDNA2 will see a roughly
similar 1440p->4k scaling as RDNA1 does. This bit is what’s most likely to
be sketchy, but there’s no way to evaluate it until at least the 28th,
and maybe not till release and reviews are out.
Big Navi 4k performance
We don’t really know how Big Navi will perform. Most news sites like Techpowerup or Rockpapershotgun are just trying to compare their benchmarks against AMD’s. There are several issues with doing this.
-
We really have no idea what sort of test runs
AMD did. It seems that for the Borderlands 3 run, they used the in-game benchmark.
Techspowerup found that 61 was much lower than the 70fps that they got with the
3080, but they didn’t use the benchmark run and did their own run of a repeatable
section, which may have been more or less demanding, and others are finding that
by running the benchmark, they’re getting much closer to that 61 mark with the
3080 with scores around 59-62. If you look at the Call of Duty Modern Warfare
results, it seems like the 88 touted by Big Navi is FAR behind the 3080, and barely
even higher than the 2080ti. However, there’s no benchmark there. We have no
idea what area they may have gone trough, and it seems that it may have been a particularly
hardware intensive section.
-
Some games tend to favor certain hardware. The
reason that reviewers don’t just review a few games is that some games run
better on some hardware than others. As Hardware Unboxed has shown us many times,
some deltas between GPUs will be much higher or lower in various individual
games than the average will show. Borderlands 3 currently seems to slightly
favor Nvidia (by about 3-4%), so it would seem that if that’s any indication, 3080
being faster than Big Navi by a smaller margin than that, may indicate that Big
Navi is faster on average. However, looking at Gears 5, the opposite story is
told; the game tends to historically favor Radeon on average but in this case
the comparisons tend to show that Gears 5 runs better on the 3080 than Big Navi,
which makes it impossible to get a clear reading of Big Navi performance based
on historical trends.
At this point, I’m forced to simply toss out a rough
estimation of what I expect Big Navi performance to be compared to the 3080 at
4k. Simply for the sake of argument, I’m going to say that in those same 14
games that Hardware Unboxed used to find their averages, that they will get a
4k average of 100. I say this because it’s a nice clean easy number, and is if
anything, slightly lowballing. It could be higher; it could be lower. It’s a
guess. Don’t crucify me if I’m wrong.
Calculating 1440p performance:
This assumption, as mentioned, is that RDNA2 will scale similarly
from 1440p to 4k as RNDA1 did, which is to say, not that well, or at least
worst than Ampere, or even Turing which scaled worse than Ampere did. I would
say that if you want to get the most out of your 5700XT, you should play at
1440p, and if you want to get the most out of your 3080, you should play at 4k.
(You really should play at the resolutions you prefer, don’t take me too
seriously)
Here is some quick analysis done on the data given above from
Hardware Unboxed. Turing scaled slightly better at 4k than RDNA did, but Ampere
scales amazingly well. It’s been speculated that this is because of the 128
cores per SM in Ampere that are difficult to keep loaded, but indeed higher resolutions
tend to load up the cores more effectively. Whatever the reason, it’s well
established that 3080 stretches its legs better at 4k than 1440p, at least as
compared to the other architectures seen.
Now, if we take that 4k performance that I guessed (really
please don’t read too much into my average 100fps; for all I know, Big Navi
beats 3080. I’m just throwing out a number to make a point about better 1440p
scaling) of 100fps average at 4k, how fast would Big Navi be at 1440p vs the 3080?
192fps
@1440p.
Yes, that’s right. If RDNA2 scales like RDNA1, even if it
gets beat handily at 4k by 8% by the 3080, it’s going to edge out even the 3090
at 1440p. If Big Navi ends up being on average the same as the 3080, it will be
12% faster than the 3090 at 1440p, and 240Hz 1440p monitors will be justified.
But who cares about 1440p?
Well, it’s true, that most people only look at 4k gaming
performance when deciding on a GPU, even if they’re actually playing at 1440p.
I hope I’ve demonstrated why this practice is dumb; different GPU architectures
scale differently. Look for performance in the resolution you actually plan on
using. After all, most people spending $500-800 on a GPU are running 1440p,
likely high framerate gaming monitors, with the minority running 4k, and even
fewer running 4k high refresh.
Yes, some people can tell the difference between 4k and
1440p, but some people prefer the higher framerates, and also can tell when
settings get lowered enough to run 4k high refresh on a decent card. In a lot
of cases, there’s not a lot of difference between high and insane graphical
settings, but low or medium on some games do take a toll on visuals.
Ultimately, it’s up to you and your preferences to decide which matters most to
you between visual settings, resolution, and framerate. In general, you’ll have
to prioritize away from one of them even with a top card.
What about Bigger Navi? Can AMD beat Nvidia at 4k?
I don’t know. Wish I did. My best speculation is that AMD
showed us their RDNA2 flagship card that’s intended to compete with the 3080.
However, like how Nvidia spent most of their presentation on their flagship 3080,
AMD may have a Halo product intended to compete with the 3090 that’s faster than
what was shown. However, like much of my speculation, it may be wrong, so take
that with a grain of salt.
I just hope that if I’m right about resolution scaling, Lisa
Su doesn’t get caught up in 4k benchmarks that are close to the 3080 and forget
that at the most common resolution for the target demographic, they’re actually
in the lead.
Comments
Post a Comment