GTX 960 2GB vs. 4GB in 2019 – Did It End Up Mattering?
GTX 960 2GB vs. 4GB in 2019 – Did It End Up Mattering?
2019-04-29
one of our oldest and most popular
videos talks about the gtx 960 4
gigabyte vs. gtx 960 2 gigabyte cards
and the value of choosing one over the
other the discussion continues today but
it's more focused on 3 gigabyte verse 6
gigabyte comparisons or 4 gigabyte verse
8 gigabyte comparisons now looking back
at 2015's gtx 960 we are revisiting it
with locked frequencies to compare
memory capacities and how its aged the
goal is to look at both frame rate and
image quality to determine how well the
2 gigabyte card has AIDS versus how well
the 4 gigabyte 960 has aged before that
this video is brought to you by MSI's
r-tx 2070 gaming z8 gigabyte card the RT
x 2070 gaming z uses MSI's dual fan
design with large blades which we've
previously tested to have him on the
best noise normalised thermal results in
the class msi is 20 70 gaming Z has a
fat heatsink furthering the focus on
reduced noise levels by allowing the
fans to spin slower RGB LEDs naturally
are abundant on the card but can be
blacked out to match the carbon and
blackout shroud learn more at the link
in the description below first of all a
lot of things have changed since that
content number one we've changed a lot
our testing methodology has improved
further in the obvious four years since
then so that's a topic on its own where
results today will be different from
results four years ago further the
environment has changed Windows 10 has
changed significantly I think we were
using tenets that we probably use 10 of
time I don't know whatever was out I
don't know if 7 or 10 at this point or 8
maybe it might have been 8 but either
way the OS has changed significantly and
also drivers have changed things like
that so we haven't normalized for those
variables we're not doing now vs. then
what we're doing isn't now so 2 gigabyte
first for gigabyte today and we've done
a few more steps here than we might have
in years past to further improve the
accuracy of the data and one of those is
to equalize these in frequency
fortunately because the 900 series this
is pretty easy to do it's Boost 2.0 it's
before Nvidia really started pushing the
thermal
pendant frequency that we see in modern
architectures and so it's easy to lock
the to the same exact frequency down to
the megahertz one of them and same
memory frequency as well so everything's
identical here the cooling was maxed out
on both cards but it's relevant because
boost doesn't behave the way it does
today and that allows us to get a strict
a B comparison between four and two
gigabytes even though it's two different
cards they're clocked the same so it
actually doesn't matter that they have a
different heatsink and fat so locked
frequency is a big step for ensuring the
data is accurate another thing that we
need to talk about is memory used versus
memory allocated or requested a lot of
people talk about how much memory is
being used by an application when they
open gpu-z or task manager or an on
screen display and what that is showing
you is how much memory the application
has requested and that doesn't mean
what's actually engaged or what's
actually needed to run the program so
some applications might see 11 gigabytes
on a card and request all 11 gigabytes
but in reality they might only be using
4 and so you really can't rely on that
number to mean anything it doesn't
practically tell us anything we have to
do image quality test and frame rate
tasks for frame time tests to see if
there's any difference in the memory
capacity it's not enough to just look at
how much is used by a game because a lot
of them will request more than they
actually need actively and finally as we
get into the results here an important
thing to think about and remember is
that as we do things like increase the
resolution increase the texture quality
we will definitely exceed the 2 gigabyte
limit of this card especially when we're
using modern titles so because of that
we are going to see differences in
performance as it becomes limited in
memory again especially with modern
titles but when we're looking at a game
running at 26 FPS versus 30 although
that is percentage-wise a significant
change in terms of playability it
doesn't matter you're not really going
to be happy with either experience so
that's that's something to think about
as well is even though we can see at
damn it what the differences are between
the cards if we force it you still have
to question practically whether it
matters to you in real life anyway let's
get into the testing will start with
Sniper Elite Four cypher leap for is the
absolute best example to start with and
it's because the FPS number is
completely betray what's happening on
the screen if we look at our GPU bench
the chart starts at a 1080p for this one
in Sniper Elite 4 you'll see that the
gtx 960 strix two gigabyte card actually
posts a 57 FPS average with a low as a
well spaced at 41 fps and 39 FPS
comparatively the 960 SSC for gigabyte
card is within margin of error at 58 FPS
average lows are also within our wider
error margins of the smaller data sets
for 1% lows as a frame rate it looks
like there's no difference and that the
4 gigabyte card might actually be quote
overkill as you could postulate that the
rest of the GPU might not be keeping up
enough for the memory capacity to ever
matter in reality we needed an image
quality comparison and this is a newer
game that treats GPUs a bit differently
than games did in 2015 all games will
handle this their own way but Sniper
Elite 4 handles vram limitations by just
silently how though obviously taking
texture resolution and quality to
compensate for over extension on vm
consumption let's start by looking at 4k
just to really exaggerate this effect
it's immediately visible even without
the side-by-side comparison that there
are big differences in image quality
again especially at 4k where we're just
stressing the card so much as
prioritizing running at that resolution
rather than running the image where we
want it so we have issues with shadows
we have issues with mesh quality with
texture quality everything is worse and
that is something that doesn't show up
in FPS data and as such FPS data is
invalid for showing this comparison this
happens at 1080p too it's to a lesser
degree and it does take a few minutes to
really start to take away that texture
quality because it takes a little bit of
time for the memory to reach capacity 2
for the game to start exceeding the
memories capacity and so we see issues
with image quality about 1080 and 4k
with 1080 being more relevant it's just
that 4k is more obvious than it happens
immediately but either
this does invalidate the frame rate
numbers because well they don't mean
anything if the image being rendered
changes it's no longer a controlled
scenario but it does illustrate the
issue with running two gigabytes on a
960 in the modern era apex legends at
1080p positions the 960 SSC for gigabyte
at a 42 FPS average when clock matched
with the Strix 2 gigabyte card running
lows in the range of 30 fps note that we
could improve performance with lowered
settings clearly but the goal is more to
focus on the head-to-head comparison not
to make it playable the 4 gigabyte card
leads outside of margin of error
claiming a 10% gain over the 2 gigabyte
card GTA 5 was released after the 9:00
60s it launched by just a few months so
it's the closest to launch comparison at
1080p the 962 gigabyte card ends up at
50 FPS average and as well within error
margins when compared to the 960 SSC
these are about as close as you can get
at 1440p we see the same results the two
cards are within margin of error of each
other and so we can declare that they
are functionally the same in this test
at least that's true with regard to
performance image quality also doesn't
show much of a change so unlike Sniper
Elite 4 we see here that the image
quality has the same texture quality we
see that the mesh quality is about the
same everything is the same so this is
all logical if we saw a massive changes
in the image quality we would also
expect the frame rate to go up on the
card that is reducing the image quality
f1 2018 at 1080p positions the 960 SSC
for gigabyte card at 45 FPS average one
frequency locked with a low is at 28 fps
and 16 fps 1% in 0.1 percent the 960
Strix 2 gigabyte card did end up about
at 42 FPS average enabling the 4
gigabyte card elite of 8.2%
this difference is outside of our run to
run error we did have one test pass that
had an excursion from the mean
but even if we eliminate that single
test pass the lead of the 4 gigabyte
card remains at 5.5 percent and is
outside of error margins at 1440p our
existence of the stock 960 SSC and the
clock locks 1
illustrate that the test results are
outside of error once again as these two
devices represent different test
settings run at different times and
still advantaged by the extra memory the
test range is plus or minus 0.5 FPS
average in this test for these cards the
Strix 962 gigabyte card hits 29 point 6
FPS average permitting the clock lock to
960 SSC at 1440 2 megahertz lead of
about 16 percent at 1440p this is a
substantial lead and although 1440p
isn't a particularly good experience
with either of these cards the important
part is that it stresses vram and shows
us the limitations these are further
illustrated with the significantly lower
1% and 0.1% lows on the 2 gigabyte card
demonstrating frame time variance and
inconsistency even at 1440 PF 1 2018
image quality remains the same between
the 2 gigabyte and 4 gigabyte devices we
drop performance from swapping in and
out of video memory more aggressively
but we don't seem to have a reduction in
image quality far cry 5 @ 1080p doesn't
illustrate differences of any major
margin we are at 44 FPS average for the
964 gigabyte variants with the Strix 962
gigabyte card at 40 2.1 FPS average
these are nearing error for this title
although the range technically does exit
it just barely at 1440p the difference
is emergent with greater resolution the
964 gigabyte cards maintain a lead of
about 11% since each frame at this rate
is meaningful and the 0.1% lows also
show some gains on the 4 gigabyte card
far cry 5 also doesn't suffer an image
quality reduction which is it largely
demonstrated by the frame rate
difference in performance we'd likely
see an image quality reduction
accompanied with the frame rate increase
which is not observed here shadow of the
Tomb Raider at 1080p places the
frequency loss to 960 at 36.7 FPS
average allowing it a lead of about 14%
every single task pass for the 964
gigabyte was between 36 point six and
thirty six point two eight FPS average
marking this benchmark as highly
reliable and with minimal variance the
960 Strix demonstrated more variance
with results between 31 point 7 and 33
point 1 FPS average 1440p again shows
the same scaling that we've seen before
with
for gigabyte card pulling ahead in a
measurable way but again we're at a
point where you wouldn't be playing on
either card anyway it's still important
to show the difference just not
particularly meaningful to the user
experience shadow of the tomb Raider's
image quality also looks the same
between the two some characters are
randomly generated but looking at floor
tiles we see the same texture quality
and resolution the same is true for Lara
Croft who has the same texture and mesh
quality in each scene and the tree in
the poll behind her also have the same
image quality so as always with this
type of thin the answer is well it sort
of depends it's on average the four
gigabyte card is definitely better and
we do have a serious issue in Sniper
Elite 4 where the game treats this
exceeding of the memory capacity by
lowering the quality of the visual as
you've said so your your quality
settings actually mean nothing at that
point and clearly the four gigabyte card
today is doing better now the 4 gigabyte
card in 2015 versus the 2 gigabyte card
in 2015 we test it at the time and it
really again depended on the title but
the differences are not always present
but they are sometimes present which is
annoying as a conclusion because it
really just means that you would have
had to look at the types of games you're
playing at the time versus what was
tested online and see if it mattered to
you but just like today where we look at
four versus eight and we say well yeah
it's better on a twenty eighty then four
would be it's going to come down to the
GPU that's accompanied with the memory
and to some extent when you actually
play with these things at reasonable
settings they're really not all that
different the differences start to
emerge once you start beating them up so
hard with texture resolution with screen
resolution with the mesh quality things
like that that the memory is is finally
becoming a limitation but by the time
you've done that also the GP is a
limitation and it's not really a fun
experience anyway so there's a bit of a
balance between academic exercise and
real-world experience however we can
clearly see that the two gigabyte card
is doing worse
in fact with modern games not really a
huge surprise but if you're still
running a 960 you might be able to get a
little bit further with a four gigabyte
model today than you would with a two
gigabyte model today and still stretch
it out just just that much longer
through games as they launch without
needing an upgrade pathway so anyway
that's the two versus four gigabyte in
2019 there are differences the image
quality is the most interesting one and
also tells us a story about how in
testing we need to pay attention to
things like image quality and not just
the numbers because clearly some games
resolve this differently than others and
that can invalidate test data all
together if not accounted for so that's
it for this one thank you for watching
as always you can subscribe for more or
go to stored I Cameron's access net to
help us out directly by picking up one
of these shirts which we've just
restocked
it's the blue print shirt I'll see you
all next time
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.