NVIDIA GTX Titan X (Pascal) Review vs. GTX 1080 , SLI 1070s
NVIDIA GTX Titan X (Pascal) Review vs. GTX 1080 , SLI 1070s
2016-10-26
it's been about a month since we tore
down Sam's Titan xp for our
liquid-cooled hybrid build and returning
today to fully review the Titan ax
Pascal as a gaming device our first
content looked at thermal and framerate
performance when under the stock cooler
versus our custom liquid solution
proving that the Titan XP improves its
clock rate by an average of about 200
megahertz just by using a better cooler
that's no overclock necessary by the way
it's just better cooling solution
letting us better clock rates and thus a
better frame rate at least marginally
but today we're properly going over the
thermals the frame rate in various games
that we didn't cover previously for the
Titan X Pascal card and then we'll also
talk overclocking and look at it as a
standalone solution without our liquid
cooling help to see if it's worth your
purchase before getting to that this
Titan XP coverage is brought to you by
an tech and their new cube mini ITX case
designed by Razer supports full custom
loop liquid cooling in an SFF box you
can find more information on this case
in the link below the GTX Titan X Pascal
is built more for production it seems
like than gaming but that hasn't stopped
Nvidia from haloing the card as the best
for gaming out there and it is a $1000
plus card depending on where you buy it
and what time of day it is I guess and
today we're looking at the temperature
and gaming performance things like that
we're not looking at production
workloads of CUDA acceleration premiere
or anything like that it's strictly
gaming because that is how it has been
marketed pretty heavily lately so we're
gonna review it from a gaming
standpoint and this car just as a
reminder was a loaner provided back by
one of our viewers and readers so thank
you for that Sam this was not an Nvidia
sample because I guess I don't know they
didn't want to sample us a Titan XP so
this card the XP uses a GP 102 GPU that
is a Pascal architecture GPU on the
board GP 102 it's the largest Pascal
chip presently available on a geforce
card it's got more GP sees more SMS all
of that stuff we'll talk about that
momentarily the only larger Pascal chip
presently available is the GP 100 which
is only available on the Tesla P 100
accelerator that is not a gaming card by
any stretch of the imagination
that is a deep learning and simulation
card so very different market GP 102
though is the top of the line at least
in the gaming space GP 102 hosts a total
of six G pcs for 28 SMS and 3584 to two
cores at 128 cores per SM so if you
wanna to follows GP 104 is architecture
splitting gbc's mostly into sets of 5
SMS rather than following the dense or
GP 100's M alignment of 10 multi
processors per G PC we talked about that
in our GP 100 deep dive article and
video this architectural decision is
because the GP 102 chip isn't meant to
handle the same FP 64 double precision
and FP 16 half precision tasks that
you'd find on a Tesla P 100 accelerator
which again is built for simulation and
deep learning applications instead GP
102 focuses it scores on FP 32 useful
for gaming and production applications
almost exclusively also like GP 104 -
400 the Titan axpy runs 8 TM use per SM
totaling 224 texture map units clock
rate natively operates at fifteen thirty
one megahertz with the stock cooler
often hitting seventeen hundred
megahertz or higher during load note
also that we pushed it to nearly 2,000
megahertz in our liquid cooled content
coverage stuff and that was with no
overclock just as a reminder so hitting
nearly 2k on the frequency range just by
adding liquid and memory of course
operates at 10 gigabits per second with
gddr5 X from micron that is not news to
anyone cache size also plays a big role
in the Titan XP it's expanded to 30 72
kilobytes over 2048 on the GTX 1080 and
the RAM on board is 12 gigabytes with
Titan XP versus 8 gigabytes for the GTX
1080 both using the same g5 acts ten
gigabit per second native memory if this
architecture discussion interests you
you want to learn more about what these
terms mean and hit the article linked in
the description below where we go into
all of it and define better what the
acronyms and initialisms stand for let's
move on to testing note that all
thermals fps noise and overclocking
tests were performed before our initial
teardown of the GPU so the only
difference you'll see in numbers emerges
after we apply to a liquid cooler to the
Titan XP but all the initial numbers
acted prior we've also already produced
that content but we'll revisit the
thermals briefly here for the rest of
the testing methodology and for
additional tests again hit that article
in the description below recapping the
basics of the thermal results the
reference GTX Titan
X which is basically the only card
design hits its thermal limit at 83 to
84 Celsius GPU diode temperature non
Delta and that's what creates the boost
specification of 1531 megahertz the
clock actually automatically boosts
higher through GPU Boo's 3.0 when that
thermal limit is bypassed by superior
cooling again explained elsewhere and
our hybrid mod for instance brings us
down to 19 point 8 5 Celsius delta T
from 59.4 on the reference cooler which
uses just a vapor chamber and blower fan
that's a reduction of 40 C for the load
temperature as for idle temperatures
those are a little lower than the 1080
for a few reasons if it confuses you one
the die size of the Titan XP is 471
millimeters squared on the GPU 102 chip
whereas the GP 104 trip is 314
millimeters squared this extra service
area on GP 102 does actually help
dissipate heat in a substantial way and
then for liquid testing we also improved
our implementation by keeping the base
plate on the Titan X hybrid not done for
the 1080 hybrid keep in mind also that
these results are for the out of box
product so there's not even overclocking
and we've already dropped the thermals
by 4 TC you can learn more about the
endurance performance and other thermal
tests in the article links below
there are a couple of AIB partner card
Titan X's out there that we don't have
aces got one EVGA is making a few but
this is a quick look at what the
reference card offers we tested GTA 5 at
4k with very high in ultra settings but
threw in a 1080p test just to
demonstrate a point at 1080p were
clearly hitting a CPU bottleneck and
other component bottlenecks that's
limiting GPU performance Titan XP is
operating at 136 FPS average with 95 and
85 FPS 1% 0.1% Lewis perspectively which
is trailed just barely by the GTX 1080
suite that we've got on the bench if
you've got plans to buy either device
between the two are comparing here it's
probably worthwhile to invest in a
higher resolution to screen just because
really you shouldn't be playing 1080
with these cards that point made let's
move on to 4k test
the gtx titan x is outclassed by sli gtx
1070 s in this benchmark which jointly
post a 79 FPS average against the titan
x is 73 FPS average the single titan x
as is often the case with the multi-gpu
configurations outputs better frame
times than the sli cards and even still
the performance metrics are high enough
on each device to be more or less
identical with regard to use a
perception of framerate throughput the
difference between the configurations is
about 8% just for reference and the GT X
1080 a IV partner cards meanwhile
operate at 59 FPS average or there
abouts with a gap of approximately 23%
from the Titan XP below these the
crossfire rx 480 is rest at 58 FPS
average with a GTX 1080 Fe at 56 FPS
average we tested doom with both OpenGL
and Vulcan the latter of which is
presented as an average and comparative
charts let's start with just the OpenGL
results at 1440p with ultra settings the
Titan XP is posting an average FPS of
138 coupled with nearly 100 FPS 1% lows
and about 89 fps 0.1% lows the Gigabyte
GT X 1080 extreme water forest card that
we reviewed recently is next in line and
that's at 128 FPS average I'm similar
lows it's not until we get to the 1080
EFI that there's a reasonable gap
created almost entirely by the clock
rate difference between all the cards
where the 1080 Fe pushes 109 FPS average
for Vulcan at 1440p the comparison shows
an FPS output of 155 average for the
Titan XP and that's followed by a gtx
1080 at about 128 FPS average that's
performance difference of roughly 21% or
35 percent for the EFI variant and we
need to tax the cards a little harder to
show any visible degradation and
performance so we're moving on to 4k
with OpenGL only we've only got a few
cards present on the bench for doom
testing and the GTX Titan XP pushes an
average FPS of 81 with lows north of 60
fps the GTX 1080 reference card is
performing at 60 FPS average with 51 FPS
1% lows and 49 fps 0.1% low metrics
Volken posts the Titan XP and DT X 1080
at roughly the same perform
output since we're becoming bound by the
resolution more than anything
moving on to Mirror's Edge catalyst at
1440p stock tighten XP performance runs
at approximately 111 FPS average with
low frame rates reasonably timed to the
average and constantly above 60fps the
gigabyte GT X 1080 extreme water forest
card along with all the other gtx 980 is
rests at 95 FPS average and the
reference 1080 is at 89 or about 24%
slower than Titan XP moving to 4k the
Titan X Pascal card is able to sustain
high quality settings at 3840 by 2160
with an FPS of 60.3 FPS average about 4
to 5 ahead of the liquid-cooled 1080
GPUs and 5 FPS ahead of SLI GTX 10 70s
the Seahawk and 1080 gaming X both run
at around 51 FPS average with a 1070
gaming X way down at 40.6 7 average and
30 fps lows for more charts including
Metro last light and Mordor in a couple
of other games hit the link the
description below for now let's move on
to overclocking this is our overclocked
stepping table with the GTX Titan XP
card with maxed out power target and no
core offset if you look closely in the
top few rows the average clock rate
operates at about a hundred megahertz
higher than stock with a 100% target but
we're still thermally limited regardless
the maximum stable overclock finalized
at around 1911 megahertz average
frequency with a 175 megahertz offset to
core and 450 offset to memory the peak
clock was 1920 3.5 megahertz
briefly anyway and this was with a 3500
rpm fan speed so noise levels are
getting a bit more significant the
liquid cooled variant for the curious
was able to sustain a maximum overclock
of 2012 megahertz where 1974 megahertz
average and here's a quick look at
overclocking performance we're seeing
fairly substantial gains in some titles
nearing 10 FPS more in GTA 5 and
Mirror's Edge catalyst just to sort of
reiterate our hybrid research content
for the Titan XP this card and video is
technically operating within spec
because the spec says 15 31 megahertz
it's often boosting to 1700 plus in
gaming but the thin is again the spec
could be higher and it doesn't
retuning the GPU or using a different GP
100 whatever processor it just requires
a better cooler putting on this
reference cooler although they say it's
high quality and really it's not bad as
far as reference coolers go but putting
it on this card seems sort of wasteful
because you can get an extra 3 point 5
to 5 percent performance out of it just
by using a better cooled card and that
can be done with liquid or it could be
done with air and the thing is because
of the Titan X is sort of an Nvidia
exclusive model it's really difficult to
get AIB partner cards there aren't many
of them I know
EB J's got a hybrid model they're making
or have made and there might be a couple
others in the works but there aren't a
lot of them so that means the most
common model will be the reference Titan
XP and that's just it's basically again
a 5 percent performance swing just for
using a better cooler so that's not
something I was too happy about you
could sort of bypass that and I'm sure
some folks would argue that you could
run the vrm fan at 100% and you'd get
that extra bit of performance but that's
really loud and it's just not worth it
the Titan XP is still priced north of
$1,000 for the most part with the GT X
1080 resting closer to 700 these days
and it's absolute best performing
scenarios the Titan XP is able to
outperform the 1080 EFI by roughly 30%
again best case scenarios and in the
best case also post gains over a B
partner 1080 s upwards of about 25% but
we've got to keep the bigger picture in
mind a GTX 1080 is already capable of
running almost every game we've tested
at 4k with roughly 60 FPS frame rates at
a minimum in some cases and for most
enthusiasts at the high end we'd wager
that's enough and extra $300 doesn't
really gain tangible frame rate
improvement at this point since we're
already so high at fps output up 1440p
and in some 4 case scenarios though not
all now that's not always true the Titan
XP does have a few specific games where
its performance gains are substantial
but for the most part the GTX 1080 makes
more sense as a top-of-the-line gaming
card and is better priced for the sort
of average enthusiast user Titan XP it
may make more sense for rendering cuda
accelerated applications but we're still
waiting on more
those to support Pascal architecture
before really diving into testing that
some things like blender have only just
started adding support for Pascal so
it's still a little ways off in terms of
our test suite for games though the 1080
is pretty good as a top-of-the-line card
I don't really know that there's a needs
you go to the Titan XP unless you've got
some really specific use case in which
case you probably know if you already
need it or not one last thing to throw
in here is the name so if we talk about
this with the gtx 1060 it's got ten
percent fewer SMS half the vram and it's
called a 1063 gigabyte but there's a
1066 gigabyte it's got more as Hatton's
twice the ram i can kind of get over the
vram difference but the sm change alone
should be enough to issue a kind of
rebrand of some kind of rebadge Titan XP
is not the same problem because the spec
there's one spec there's one SKU
it's just Titan X is the name I've been
calling it XP or Titan X Pascal this
whole time that's because the Titan X mm
which used to be called the Titan X was
the Maxwell card and if you go look
anywhere online you type in Titan X on
Newegg almost all of the results are
going to be for Maxwell cards if not all
of them and they're priced similarly so
it'd be pretty easy to accidentally buy
an ace used Titan X Maxwell card which
performs pretty differently a different
architecture and that's something to
look out for and it's something that
really I wish Nvidia had enough
foresight to to not do that it would
literally adding one letter to the name
would have fixed that issue Titan XP and
it doesn't even sound bad it's cut the
gaming sound to it with XP so if you are
going to buy one of these which it's
good card I'm not saying it's not a good
card it's just it's very high performing
it may be not necessary for a lot of
people but if you're gonna buy one be
careful when you look around on ebay
Newegg things like that make sure you're
not getting a maxwell card go directly
to Nvidia's website if you want to be
really safe because that's where you'll
definitely get a pass go on so that is
all for this time is always patreon link
or post your video if you like this type
of coverage and want to help us with
more and then subscribe for additional
content coming out as a ways everyday
for the most part links to the
description below for more 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.