WEBVTT

00:00:00.320 --> 00:00:09.519
let's begin with the important stuff the v in titan v isn't a roman numeral there

00:00:07.279 --> 00:00:15.040
have actually been six titans now making this one number seven

00:00:11.519 --> 00:00:18.400
instead it directly signifies the volta

00:00:15.040 --> 00:00:20.480
architecture that this card is based on

00:00:18.400 --> 00:00:26.000
and to further drive home the point that it is not a gaming oriented card NVIDIA

00:00:23.680 --> 00:00:33.120
has decided to forego the GeForce gtx in the product name not

00:00:30.560 --> 00:00:38.160
that we won't game on it of course after i tell you guys about ifixit their

00:00:36.160 --> 00:00:42.719
protec toolkit gives you the tools you need to tackle any electronics repair

00:00:40.239 --> 00:00:54.559
challenge visit ifixit.com forward slash Linus at the link below to learn more

00:00:54.559 --> 00:01:00.079
so to help manage overhead our test setup is a little different this time

00:00:58.320 --> 00:01:05.920
around we're going to be testing our productivity and our gaming loads on our

00:01:03.120 --> 00:01:11.040
z270 bench for its fast per thread performance while we'll be doing our

00:01:07.680 --> 00:01:12.880
scientific testing on our x299 bench

00:01:11.040 --> 00:01:17.680
where those workloads are more likely to benefit from the extra CPU cores

00:01:15.920 --> 00:01:25.600
let's kick things off then with the boring sciency stuff i'm just kidding so

00:01:22.080 --> 00:01:28.080
titan v manages a higher

00:01:25.600 --> 00:01:34.640
percentile frame rate in deus ex mankind divided than the second place titan xps

00:01:31.040 --> 00:01:37.759
in directx 11 mode then it continues the

00:01:34.640 --> 00:01:40.560
curb stompage in directx 12. middle

00:01:37.759 --> 00:01:46.079
earth shadow of war also sees a huge spike in performance over the titan xp

00:01:43.280 --> 00:01:51.200
and while assassin's creed origins dials things back a little the titan v still

00:01:48.880 --> 00:01:57.280
ends up on top so i don't even really need to say it but given these scores

00:01:54.560 --> 00:02:02.719
and our synthetic results titan v is right now hands down the fastest card on

00:02:00.479 --> 00:02:08.800
the market for gaming and it isn't even supposed to be

00:02:05.040 --> 00:02:11.760
what sets this titan apart is its fp16

00:02:08.800 --> 00:02:16.959
and fp64 performance along with the inclusion of NVIDIA's new tensor cores

00:02:14.879 --> 00:02:22.720
which are special compute units that are optimized for deep learning and ai

00:02:20.239 --> 00:02:27.520
oh and while we're at it NVIDIA is also showing off hbm2 memory for the first

00:02:25.840 --> 00:02:32.239
time in a titan so v does end up with a smaller frame

00:02:29.920 --> 00:02:39.680
buffer than the vega workstation cards or the hbm2 equipped quadro gp100

00:02:36.400 --> 00:02:44.920
but it may end up making up for that

00:02:39.680 --> 00:02:48.239
through sheer brute force its whopping

00:02:44.920 --> 00:02:51.840
5120 cuda cores should make it the

00:02:48.239 --> 00:02:53.280
fastest card available for compute

00:02:51.840 --> 00:02:56.959
should but when we first took the shrink wrap

00:02:55.360 --> 00:03:01.840
off and loaded up some compute benchmarks we actually found that many

00:02:59.840 --> 00:03:07.360
cuda focused tasks simply wouldn't run well it turns out

00:03:05.360 --> 00:03:12.319
that while previous generation cards were built around cuda compatibility

00:03:09.200 --> 00:03:15.360
version 6.1 and earlier

00:03:12.319 --> 00:03:18.800
volta and by extension the titan v bumps

00:03:15.360 --> 00:03:21.120
this up to 7.0 and new compute kernels

00:03:18.800 --> 00:03:26.640
need to be compiled to support it so blender most coin mining algorithms and

00:03:24.319 --> 00:03:30.799
other tests were affected with users reporting even more incompatibilities

00:03:29.120 --> 00:03:34.720
that we didn't see oh i'm getting flashbacks to the vega

00:03:33.040 --> 00:03:39.599
frontier edition thankfully the opencl layer still works

00:03:37.680 --> 00:03:45.519
though so we can get rough numbers for benchmarks that support NVIDIA cards in

00:03:42.319 --> 00:03:48.239
opencl mode with the caveat that opencl

00:03:45.519 --> 00:03:52.959
does have some overhead compared to cuda starting off with luxmark we've got

00:03:51.920 --> 00:03:58.640
wow are these numbers even right

00:03:55.360 --> 00:04:01.599
luxmark is super paralyzed so those

00:03:58.640 --> 00:04:08.799
extra cuda cores pulled double the performance of the titan xp and

00:04:04.799 --> 00:04:11.200
far higher than even the quadro gp100

00:04:08.799 --> 00:04:16.720
chaos group's render suite v-ray brings us another win for the titan v shaving a

00:04:13.840 --> 00:04:21.440
cool three seconds off the gp 100 in second place moving on to compubench we

00:04:19.519 --> 00:04:25.919
see the gap close for the level set segmentation benchmarks but then the

00:04:23.840 --> 00:04:31.040
titan v pulls way out ahead in the computer vision optical flow benchmark

00:04:28.479 --> 00:04:36.720
leaving everything else well in the dust and it continues to assert its dominance

00:04:33.440 --> 00:04:40.800
in most of spec view perf where 3ds max

00:04:36.720 --> 00:04:43.280
kittia maya energy and showcase all have

00:04:40.800 --> 00:04:47.520
it at the top but it should be noted that the quadros do fire back in some of

00:04:45.360 --> 00:04:52.639
the other tests thanks to their enhanced driver code paths

00:04:49.280 --> 00:04:53.520
as for coin mining well the rumors are

00:04:52.639 --> 00:05:00.160
true it is a beast especially an eighth hash

00:04:58.000 --> 00:05:06.000
but with that said the three thousand dollar price tag is going to kill your

00:05:03.040 --> 00:05:12.400
roi time and it doesn't even win in every scenario vega frontier handily

00:05:08.479 --> 00:05:12.400
outperforms it in kryptonite

00:05:12.720 --> 00:05:20.320
that's a lot of testing and it looks like volta is mostly living

00:05:17.039 --> 00:05:22.000
up to its promises so far but we've got

00:05:20.320 --> 00:05:28.320
one more test up our sleeves for its compute performance deep learning

00:05:24.880 --> 00:05:31.360
using imagenet 12 on caf2 titan v

00:05:28.320 --> 00:05:33.520
managed a significant lead at both

00:05:31.360 --> 00:05:36.720
tested batch size thanks to its tensor cores

00:05:34.400 --> 00:05:43.520
but maybe even more interestingly it even wins without them enabled

00:05:40.320 --> 00:05:46.400
against the gp 100 a truly monstrous

00:05:43.520 --> 00:05:52.880
feat because for context you guys that card is still almost triple the

00:05:49.759 --> 00:05:55.039
price well okay then lioness fine

00:05:52.880 --> 00:05:58.880
volta is great for compute big whoop what does that mean for me

00:05:57.360 --> 00:06:02.400
okay i'll get to that first we need to talk about thermals and

00:06:00.560 --> 00:06:09.840
power because this bit is really important while titan v manages to throttle a bit

00:06:06.400 --> 00:06:13.039
harder than the gp100 and the gtx 1080

00:06:09.840 --> 00:06:15.680
ti in our testing and while it has the

00:06:13.039 --> 00:06:19.840
second highest idle power draw in our entire test lineup

00:06:17.680 --> 00:06:23.759
thanks at least partially to NVIDIA using tsmc's 12 nanometer manufacturing

00:06:22.720 --> 00:06:30.000
process it's also got the lowest full load power

00:06:27.280 --> 00:06:35.120
draw which is bonkers considering everything we know about its overall

00:06:32.840 --> 00:06:41.280
performance so then back to your question what this means for you is that

00:06:37.680 --> 00:06:43.680
the slightly cut down and maybe

00:06:41.280 --> 00:06:49.440
higher clocked enthusiast consumer variants without any tensor cores are

00:06:46.400 --> 00:06:51.199
going to be mind-bogglingly fast

00:06:49.440 --> 00:06:57.280
but that's a conversation for another day here and now what we have is a

00:06:54.240 --> 00:07:00.319
quadro killer at a third of the price

00:06:57.280 --> 00:07:02.560
for most workloads and one that can game

00:07:00.319 --> 00:07:05.360
faster than anything that's ever come before it

00:07:03.919 --> 00:07:11.360
though i'm not saying you should go out and buy one if raw horsepower for compute is your

00:07:08.560 --> 00:07:16.960
jam and you won't benefit from a bigger frame buffer or multi GPU with nvlink

00:07:14.880 --> 00:07:19.440
these connectors they're dead you're looking at your next card right

00:07:18.479 --> 00:07:26.080
here though you'll need to roll your own cuda kernels or wait until compatibility

00:07:23.520 --> 00:07:31.199
level 7 is mainstream by which time there will probably be a quadro variant

00:07:28.800 --> 00:07:37.199
as for the gamers well the lack of g-force in the branding says

00:07:33.680 --> 00:07:40.080
it all it is just not worth it even for

00:07:37.199 --> 00:07:42.240
you mr or ms mega baller over there i mean

00:07:40.720 --> 00:07:46.400
do you really want something that's not even sli compatible anyway

00:07:44.639 --> 00:07:49.840
so you have to run a single card in your system like some kind of a peasant

00:07:48.800 --> 00:07:54.960
so think of the v not as something to buy

00:07:52.800 --> 00:07:58.960
but as more of a teaser for what may come in 2018

00:07:57.120 --> 00:08:04.879
i know that i'm feeling sufficiently teased right about now

00:08:02.080 --> 00:08:08.479
speaking of teasing freshbooks they don't tease they give you the whole

00:08:06.639 --> 00:08:12.960
thing it's the small business accounting software that's custom built for how you

00:08:11.039 --> 00:08:16.639
want to work if you're a freelancer or a small business owner freshbooks is a

00:08:15.039 --> 00:08:20.720
simple way to be more productive more organized and get paid faster you can

00:08:19.280 --> 00:08:24.479
create and send professional looking invoices in less than 30 seconds you can

00:08:22.960 --> 00:08:28.400
set up online payments with just a couple of clicks to get paid up to four

00:08:26.160 --> 00:08:32.000
days faster you can see when your client has seen your invoice to put an end to

00:08:29.840 --> 00:08:37.039
the guessing games and they've got fully featured apps for iOS and Android so you

00:08:34.719 --> 00:08:41.360
can take all of your timesheet tracking and expense tracking with you on the go

00:08:39.760 --> 00:08:45.200
for an unrestricted 30-day free trial just go to freshbooks.com tech tips

00:08:43.519 --> 00:08:48.800
we'll have that linked below and enter Linus tech tips in the how you heard

00:08:46.880 --> 00:08:51.680
about us section so thanks for watching guys this video

00:08:50.480 --> 00:08:55.360
sucked you know what to do but it's awesome get subscribed hit that like

00:08:53.519 --> 00:08:59.600
button or maybe check out where to buy the stuff we featured at the link in the

00:08:57.839 --> 00:09:02.800
video description we should just sell ours like what are

00:09:01.120 --> 00:09:06.240
we gonna do with this beyond making this video anyway also

00:09:04.640 --> 00:09:13.480
linked in the video description is our merch storage has cool shirts like this one and our community forum which you

00:09:09.760 --> 00:09:13.480
should totally join