WEBVTT

00:00:00.240 --> 00:00:05.520
making your games look better without having to drop a bunch of money on a new

00:00:03.760 --> 00:00:10.000
graphics card is something that GPU manufacturers are keenly aware that

00:00:07.600 --> 00:00:13.759
gamers want especially now that a new GeForce or Radeon card is going to cost

00:00:12.160 --> 00:00:18.960
you a spare kidney on the secondary market now NVIDIA and AMD have both tried to

00:00:16.640 --> 00:00:24.720
use their own secret sauces to make this happen both NVIDIA's dlss and AMD's fsr

00:00:22.720 --> 00:00:28.560
have made plenty of headlines recently but team green is adding another feature

00:00:27.119 --> 00:00:34.640
to its arsenal that it hopes will give it a leg up on AMD and maybe give you a

00:00:31.199 --> 00:00:36.559
leg up in your games they call it dldsr

00:00:34.640 --> 00:00:40.719
or deep learning dynamic super resolution and if that sounds familiar

00:00:38.640 --> 00:00:44.239
it's because NVIDIA's regular dynamic super resolution you know without the

00:00:42.640 --> 00:00:50.160
deep learning part has been with us since the days of the gtx 970 and 980.

00:00:47.600 --> 00:00:55.440
so let's talk about what exactly that is and how deep learning improves upon it

00:00:53.120 --> 00:01:00.079
regular dsr is a super sampling technique basically it uses your

00:00:57.680 --> 00:01:04.799
graphics card to render your game at a higher resolution than what your monitor

00:01:01.840 --> 00:01:08.880
natively supports then it scales that down to match your screen this produces

00:01:07.439 --> 00:01:12.799
better looking images than just rendering at your display's native resolution

00:01:11.360 --> 00:01:17.840
but why well dsr more specifically works by

00:01:15.360 --> 00:01:22.159
taking samples of groups of neighboring pixels from the higher resolution

00:01:19.680 --> 00:01:26.400
rendered image and blending them into a single pixel that you see on your lower

00:01:24.159 --> 00:01:30.640
resolution display to make the resulting image more accurate this is essentially

00:01:28.799 --> 00:01:36.720
the same way that super sample anti-aliasing or ssaa works

00:01:34.000 --> 00:01:41.040
ssaa has been around for longer than dsr but games had to implement it

00:01:38.400 --> 00:01:45.280
individually dsr has the advantage of not requiring work from game developers

00:01:43.360 --> 00:01:49.119
and also puts the image through NVIDIA's own filter that takes into account

00:01:46.960 --> 00:01:53.360
additional surrounding pixels to further improve image quality dsr has a habit of

00:01:51.600 --> 00:01:56.960
making images a little softer than traditional ssaa though meaning that

00:01:55.119 --> 00:02:00.880
some gamers like it and others don't regardless it's a popular way to improve

00:01:58.960 --> 00:02:05.280
image quality especially as it allows more granularity with exactly how hard

00:02:03.200 --> 00:02:09.200
you want your GPU to work instead of rendering the image at two times or four

00:02:07.600 --> 00:02:14.720
times your monitor's resolution you could specify something like 2.25 times

00:02:12.400 --> 00:02:19.760
depending on how powerful your GPU is but both dsr and ssaa are very

00:02:18.080 --> 00:02:25.200
computationally expensive since you're basically asking your GPU to do way more

00:02:22.640 --> 00:02:30.319
work and then throw away most of it that is where dl dsr comes in it

00:02:27.760 --> 00:02:34.400
leverages the tensor cores in rtx series gpus that are designed with machine

00:02:31.920 --> 00:02:37.680
learning and ai in mind we'll tell you about how dldsr works right after we

00:02:36.239 --> 00:02:41.200
thank freshbooks for bringing you this episode freshbooks is built for

00:02:39.440 --> 00:02:44.160
freelancers and small business owners who don't have time to waste on

00:02:42.480 --> 00:02:48.319
invoicing accounting or payment processing you can save up to 11 hours a

00:02:46.480 --> 00:02:51.680
week by automating pesky admin tasks like time tracking following up on

00:02:49.920 --> 00:02:56.160
invoices and expense tracking with features like the new digital bills and

00:02:53.599 --> 00:02:59.280
receipt scanner over 24 million people have used freshbooks and it's easy to

00:02:57.840 --> 00:03:02.720
see at a glance exactly where your business stands then turn everything

00:03:00.879 --> 00:03:06.159
over to your account income tax season it comes with award-winning support so

00:03:04.319 --> 00:03:09.760
why wait try out freshbooks for free for 30 days no credit card required by going

00:03:08.319 --> 00:03:14.800
to freshbooks.com techquicky the idea behind dldsr is that

00:03:12.640 --> 00:03:19.760
the neural network and the tensor course can figure out what the image should

00:03:17.360 --> 00:03:25.599
look like by sampling a smaller number of pixels than regular dsr or ssaa one

00:03:23.760 --> 00:03:32.400
example NVIDIA gave was from the game prey which was rendered at 1620p rather

00:03:28.720 --> 00:03:34.239
than 4k then downscaled to 1080p

00:03:32.400 --> 00:03:38.799
this was accomplished with nearly no frame rate drop compared to native 1080p

00:03:36.799 --> 00:03:43.040
rendering and the image actually looked a little better than the traditional dsr

00:03:41.120 --> 00:03:48.640
image that downscaled all the way from 4k similar to regular dsr dldsr doesn't

00:03:46.799 --> 00:03:52.319
need per game support so it can actually make your older games look refreshed but

00:03:50.959 --> 00:03:56.640
if you're using a title that supports dlss you can combine the two features

00:03:54.799 --> 00:04:00.640
for an extra quality boost without sacrificing too much performance

00:03:58.799 --> 00:04:04.879
don't expect it to work miracles though early reviews have said that the quality

00:04:02.319 --> 00:04:09.439
improvement over regular dsr is pretty marginal and of course turning up the

00:04:07.439 --> 00:04:13.360
scaling factor is still going to slow your frame rate down just that the

00:04:11.280 --> 00:04:17.280
impact won't be as huge as it would be with traditional super sampling you'll

00:04:15.439 --> 00:04:21.759
need an rtx card to take advantage of dldsr but don't feel completely left out

00:04:19.759 --> 00:04:26.400
if you have a card from team red as AMD is set to launch its own ai-based Radeon

00:04:24.320 --> 00:04:30.639
super resolution technology in early 2022 which should work on any rd and a2

00:04:29.520 --> 00:04:34.560
GPU AMD cards at least for now don't have

00:04:33.040 --> 00:04:38.320
dedicated machine learning cores like NVIDIA's do so it'll be interesting to

00:04:36.400 --> 00:04:43.199
see how the two super sampling methods ultimately stack up against each other i

00:04:40.479 --> 00:04:48.720
just hope they'll both be super get it because it's super resolution

00:04:46.160 --> 00:04:51.280
super subscribe leave a like or dislike depending how you felt and leave a

00:04:49.919 --> 00:04:54.560
comment if you have a suggestion for a future fast as possible