WEBVTT

00:00:00.080 --> 00:00:06.000
it's not a new graphics card that you can't buy it's not even something that

00:00:03.600 --> 00:00:11.040
belongs to one team or the other what you're getting today yes you

00:00:08.639 --> 00:00:15.200
is an upgrade for the graphics card you already own AMD's new fidelity effects

00:00:13.440 --> 00:00:18.880
super resolution may not be the only upscaling method out there but

00:00:17.039 --> 00:00:23.119
dang it it's the only one that promises to run on graphics cards that launched

00:00:20.400 --> 00:00:25.840
as many as five years ago but should you care

00:00:24.080 --> 00:00:29.840
ridge wallet has redefined the traditional wallet with its compact

00:00:27.359 --> 00:00:33.920
frame and rfid blocking plates keep your wallet bulge down and use offer code

00:00:31.519 --> 00:00:36.320
Linus to save 10 and get free worldwide shipping

00:00:43.200 --> 00:00:47.760
we're going to start things off with image quality comparisons and we'll

00:00:46.399 --> 00:00:51.520
explain exactly how we got there afterwards but first a disclaimer while

00:00:49.600 --> 00:00:55.120
we would love to compare fidelity fx super resolution to NVIDIA's deep

00:00:53.039 --> 00:00:59.520
learning super sampling unfortunately the only title that supports both right

00:00:56.719 --> 00:01:03.760
now kings hunt is still in alpha and doesn't work on our bench yet so for

00:01:01.280 --> 00:01:07.920
today we'll focus solely on fsr we'll be using an rx-6800xt on our GPU bench for

00:01:06.159 --> 00:01:13.360
our testing though there's a small caveat of the seven games available at launch

00:01:10.880 --> 00:01:16.400
we only have these five godfall therefore will be our main point of

00:01:14.720 --> 00:01:21.200
comparison because it's both visually detailed and will let us test how fsr

00:01:18.640 --> 00:01:26.400
impacts ray tracing but first lioness is a big anno fan so i can't

00:01:23.600 --> 00:01:29.600
leave that out and keep my job can i at 4k you would be hard-pressed to

00:01:28.080 --> 00:01:32.880
distinguish the ultra quality or even quality presets from native unless you

00:01:31.600 --> 00:01:36.720
know where to look and it's good enough on balance that you'll rarely notice the difference

00:01:35.680 --> 00:01:40.799
unless you're looking for shimmering artifacts on thin or geometry the

00:01:38.960 --> 00:01:44.799
performance preset on the other hand starts to show some warts with even

00:01:42.560 --> 00:01:49.119
non-moving thin geometry visibly losing detail but for a low motion game like

00:01:47.119 --> 00:01:52.960
ano if it means you get to turn up the eye candy why wouldn't you

00:01:51.040 --> 00:01:57.680
dropping the resolution to 1440p with the fsr preset to quality or below lets

00:01:55.360 --> 00:02:00.240
some blur creep in with more significant shimmering thanks to the lower input

00:01:59.119 --> 00:02:04.880
resolution wait what do i mean by lower resolution

00:02:02.560 --> 00:02:08.239
well like dlss the lower the quality setting the lower the resolution the

00:02:06.240 --> 00:02:12.720
game actually runs at so when you set the game to 1440p ultra quality the game

00:02:10.479 --> 00:02:16.560
is actually running closer to 1080p that's why at 1080p i'm not recommending

00:02:14.720 --> 00:02:20.239
going below the ultra quality preset unless you like a lot of blur

00:02:18.319 --> 00:02:24.560
still it may be a better experience than running sub 1080p natively so pick your

00:02:22.959 --> 00:02:28.000
poison i guess onto godfall

00:02:26.160 --> 00:02:30.640
in this 4k static scene we get a good spread of edge cases to take into

00:02:29.280 --> 00:02:34.480
consideration compared to native ultra quality looks

00:02:32.480 --> 00:02:38.480
nearly indistinguishable which is awesome everything remains sharp and

00:02:36.640 --> 00:02:42.640
there's little visible loss and texture detail if any with only minor issues

00:02:40.879 --> 00:02:46.480
like the gold color detail work on this pillar the quality preset meanwhile

00:02:44.560 --> 00:02:50.239
introduces some shimmering in finer details like the self shadows and the

00:02:48.080 --> 00:02:54.080
silver means here and some details textures appear a little bit softer than

00:02:52.000 --> 00:02:57.840
they should but it's not too noticeable unless you're looking too closely the

00:02:55.840 --> 00:03:02.400
balance preset retains sharpness around hard edges remarkably well but textures

00:03:00.480 --> 00:03:06.319
are suffering much more than before and the distant stairs begin to show some

00:03:04.319 --> 00:03:10.159
unevenness in the upscale finally the performance preset using a 1080p input

00:03:08.800 --> 00:03:14.159
resolution remember gives us much blurrier textures and a

00:03:12.239 --> 00:03:17.360
more shimmery scene overall which could be distracting

00:03:15.599 --> 00:03:20.080
critically those distant stairs are now blurring together because there's not

00:03:18.720 --> 00:03:25.519
enough detail for the upscaler to properly display them hard edges with a lot of contrast are

00:03:23.040 --> 00:03:29.840
headed very well but less contrasting edges are a lot blurrier than native

00:03:28.000 --> 00:03:34.319
in motion these differences between the presets become even more noticeable

00:03:32.480 --> 00:03:38.000
now graphically simpler games like king's hunt for example are far more

00:03:36.480 --> 00:03:41.760
forgiving and the performance preset offers a surprisingly similar level of

00:03:39.760 --> 00:03:45.360
detail to native you're mostly just losing detail on distant stairs and your

00:03:43.760 --> 00:03:49.920
character's hair back to godfall here's the same scene at

00:03:47.200 --> 00:03:53.280
1440p while textures are still pretty sharp at the ultra quality setting

00:03:51.360 --> 00:03:57.040
shimmering becomes a bit more obvious on finer details as the roughly 1080p input

00:03:55.440 --> 00:04:00.400
resolution lacks the pixels needed to reconstruct them this is minor though

00:03:59.040 --> 00:04:04.560
and you really need to know where to look to pick them out quality preset on

00:04:02.400 --> 00:04:08.560
the other hand becomes a lot rougher on those finer details notice the

00:04:06.400 --> 00:04:11.920
shimmering here and the missing stairs or blurry tree leaves over here hard

00:04:10.560 --> 00:04:16.560
edges do pretty well though and if you're in the middle of a fight you probably won't notice that you're not at

00:04:14.640 --> 00:04:20.560
native revolution balanced straight off looks like an upscaled image with blurry

00:04:18.479 --> 00:04:25.759
textures and fine details getting lost look at that fine gold trim turned into

00:04:22.479 --> 00:04:26.639
a solid blob and performance preset

00:04:25.759 --> 00:04:32.240
well there's only so much you can do with an input 720p image and it's remarkable

00:04:30.240 --> 00:04:35.680
that it looks as good as it does static hard edges like the pillar itself look

00:04:33.840 --> 00:04:39.759
alright but anything that moves like the glowing pauldron on the silver main

00:04:37.360 --> 00:04:44.320
shoulder now also shimmers and fine details go out the window entirely still

00:04:42.240 --> 00:04:48.800
it beats running at 720p with monitor scaling and compared to an m cable it's

00:04:46.400 --> 00:04:51.759
noticeably sharper with more detail and you're getting it for free just like you

00:04:50.240 --> 00:04:57.360
get stickers for free with purchase at lttstore.com now 1080p is kinda scraping

00:04:54.800 --> 00:05:02.000
the barrel for fsr unfortunately with ultra quality providing okay results but

00:05:00.000 --> 00:05:05.600
being visibly blurrier and less detailed than native particularly around the

00:05:03.759 --> 00:05:09.199
sword and the pillar's trim the quality preset is as low as you

00:05:07.199 --> 00:05:14.320
probably want to go for 1080p it looks like 1440p performance mode

00:05:11.600 --> 00:05:18.720
which makes sense because like that mode it's actually using 720p for the input

00:05:16.160 --> 00:05:23.520
resolution strangely enough balance doesn't look that much worse all things

00:05:20.720 --> 00:05:26.639
considered but now even hard contrast areas like the ARM are beginning to show

00:05:25.440 --> 00:05:31.919
some fuzz and performance well

00:05:29.120 --> 00:05:36.880
it looks kind of like a 540p game on an xbox 360 or ps3

00:05:34.240 --> 00:05:40.160
and that's because it is 540p hard contrast areas are still better than

00:05:38.320 --> 00:05:44.400
that kind of upscaling would look but now to go along with a slight blur

00:05:42.160 --> 00:05:49.039
we've got some halo artifacts this isn't really how i would want to

00:05:46.639 --> 00:05:53.039
play the game but i mean if your only other option is to not play the game

00:05:51.120 --> 00:05:58.080
this is a hundred percent improvement the 720p is pushing the envelope to a

00:05:55.919 --> 00:06:02.639
frankly silly degree but it's an interesting demonstration of just how

00:05:59.919 --> 00:06:06.080
much fsr differs from dss three clicks philip made a pretty

00:06:04.080 --> 00:06:12.720
amazing video showing that going super low res with dlss can make an image look

00:06:09.280 --> 00:06:14.479
high-res but strangely blurry fsr though

00:06:12.720 --> 00:06:17.520
absolutely cannot which tells us everything we need to know about how it

00:06:15.840 --> 00:06:20.960
works without even hearing the whole explanation but i'll tell you the whole explanation

00:06:19.919 --> 00:06:28.479
anyway after we go over performance with ray tracing on at 4k our 6800xt

00:06:25.680 --> 00:06:31.520
manages a massive 40 to 46 improvement over native just by switching fsr to

00:06:30.160 --> 00:06:37.440
ultra quality going from consistently under 60 FPS to

00:06:34.639 --> 00:06:40.479
consistently well over which is a pretty great deal if you ask me considering the

00:06:39.199 --> 00:06:44.000
visual difference is basically indistinguishable unless you're really

00:06:42.400 --> 00:06:47.120
really pixel hunt we're seeing pretty linear progression

00:06:45.600 --> 00:06:54.000
down through the quality levels so depending on your pain tolerance for blurring you can squeeze quite a lot of

00:06:50.400 --> 00:06:56.240
performance out of ssr at 4k at 1440p it

00:06:54.000 --> 00:07:00.479
remains a linear improvement for a 6800 xt up until the balanced preset where

00:06:58.800 --> 00:07:03.919
the improvement is within spinning distance of quality and performance

00:07:02.400 --> 00:07:07.520
doesn't do much better still turning fsr on at ultra quality

00:07:05.919 --> 00:07:12.560
will provide a more stable frame rate for 120hz displays so that might be

00:07:10.479 --> 00:07:18.000
worth it not sure i'd go below quality at 1440p however but bear in mind that

00:07:15.280 --> 00:07:22.319
this is a 6800 xt we're talking about and older gpus that struggle at 1440p

00:07:20.319 --> 00:07:27.120
native may see better returns since it's probably the bottleneck not your CPU

00:07:25.280 --> 00:07:31.440
1080p certainly sees almost no improvement on our 6800xt with

00:07:29.520 --> 00:07:35.199
unpredictable results thanks to how CPU bound we're getting

00:07:32.639 --> 00:07:40.479
but this brings up an important question why not just lower your settings

00:07:37.759 --> 00:07:44.880
well smartypants some people don't want to and some people can't go any lower

00:07:42.800 --> 00:07:48.800
because their GPU is too weak as for why you wouldn't just run at a

00:07:46.479 --> 00:07:52.160
lower resolution and call it a day well

00:07:49.599 --> 00:07:55.680
you could but then your monitor or GPU is probably going to use bi-linear

00:07:53.599 --> 00:07:58.960
scaling which is way less sophisticated than the advanced techniques like edge

00:07:57.199 --> 00:08:03.680
directing and even neural networks deployed by the likes of dlss

00:08:01.199 --> 00:08:06.720
so why not use the best tools available and that begs the question

00:08:05.840 --> 00:08:11.759
which tool dss is great if you have the hardware it

00:08:09.440 --> 00:08:16.800
requires only found on select NVIDIA gpus of course fsr while being a little

00:08:14.800 --> 00:08:21.120
heavier to run requires no special hardware and is substantially easier to

00:08:18.960 --> 00:08:25.120
implement into a game engine easy enough that reportedly it took one developer

00:08:23.440 --> 00:08:28.879
just under two hours after being given the code to implement it into their game

00:08:27.360 --> 00:08:33.519
i mean in theory it could even be injected into a game by modders although

00:08:31.120 --> 00:08:37.519
AMD wasn't sure if it would look correct this way when we asked so

00:08:35.519 --> 00:08:41.200
your mileage may vary the effect is implemented as a shader that works by

00:08:39.200 --> 00:08:44.560
using a combination of multiple scaling and sharpening techniques that are run

00:08:42.560 --> 00:08:50.080
before the overlay effects like the hud which reduces noise and preserves text

00:08:46.880 --> 00:08:52.240
christmas christmas

00:08:50.080 --> 00:08:54.880
which reduces noise and preserves text crispness

00:08:53.519 --> 00:08:59.120
under the hood developers are given control over the scaler's variables and

00:08:57.120 --> 00:09:02.959
these can be updated on the fly no two implementations of fsr will look quite

00:09:00.959 --> 00:09:06.640
the same assuming the developer hasn't just used the defaults

00:09:04.880 --> 00:09:10.640
one thing for users to know though is that because there's a sharpening pass

00:09:08.800 --> 00:09:14.080
included with fsr having other sharpening filters enabled can

00:09:12.160 --> 00:09:17.200
ironically give the impression of more blurring than there really is thanks to

00:09:15.600 --> 00:09:21.040
the ringer effects and this is probably why our first

00:09:19.200 --> 00:09:25.120
impressions of AMD's computex reveal weren't exactly favorable

00:09:23.040 --> 00:09:29.360
unfortunately as we've seen texture blurring and fine detail shimmering is

00:09:27.200 --> 00:09:33.040
still a problem for fidelity fx super resolution today

00:09:30.800 --> 00:09:39.279
but that doesn't mean it can't change remember how bad dlss 1.0 was well bnb

00:09:36.959 --> 00:09:43.120
says we'll see new additions tweaks and new upscaling methods even in the future

00:09:41.839 --> 00:09:46.320
all of which will be open to the industry it remains to be seen whether

00:09:44.959 --> 00:09:50.959
this will work out for them or if this is another case of cuda versus opencl

00:09:48.720 --> 00:09:53.600
where NVIDIA shot first and ultimately won

00:09:52.080 --> 00:09:58.480
but developer support is pretty extensive with a lot of big names behind

00:09:55.920 --> 00:10:02.839
fsr so my money's on things working out differently this time you won't know for

00:10:00.480 --> 00:10:08.480
sure until it's been out a while but for now enjoy your free upgrade

00:10:06.480 --> 00:10:12.880
and this message from our sponsor vessy footwear thanks to bessie for sponsoring

00:10:10.640 --> 00:10:16.240
today's video vessy footwear is known for being waterproof they're lightweight

00:10:14.720 --> 00:10:20.000
and breathable and the dual climate knit material keeps your feet warm during

00:10:17.839 --> 00:10:24.320
winter and cool during the summer they fit almost any occasion and some of

00:10:21.920 --> 00:10:29.279
us are even wearing them right now and did we mention they're 100 vegan

00:10:26.800 --> 00:10:34.560
get 25 off each shoe using our code Linus tech tips at checkout check them

00:10:31.200 --> 00:10:36.800
out at bessiefootwear.com linustechtips

00:10:34.560 --> 00:10:40.640
thanks for watching guys this was a bit different from our usual videos so

00:10:38.959 --> 00:10:47.120
maybe check out our recent video on seeing if people at the office could tell whether rtx was on or not

00:10:44.880 --> 00:10:50.720
spoiler i could tell but i also knew what to look for do you