1
00:00:00,080 --> 00:00:05,520
nvidia's original g-sync technology has

2
00:00:03,280 --> 00:00:11,040
largely kept the promises that it made five years ago of eliminating tearing

3
00:00:08,480 --> 00:00:15,040
input lag and stutter in high-end gaming displays

4
00:00:12,400 --> 00:00:19,840
but the how of this technology is something that they've kept under wraps

5
00:00:17,279 --> 00:00:19,840
until now

6
00:00:22,640 --> 00:00:30,240
which is what makes today's video really really special because nvidia sponsored

7
00:00:27,760 --> 00:00:35,520
our trip down to santa clara today where we will be the first

8
00:00:32,160 --> 00:00:38,320
non-nvidia personnel to ever see one of

9
00:00:35,520 --> 00:00:41,810
their g-sync optical labs so let's go find out how you make a gaming monitor

10
00:00:39,840 --> 00:00:49,840
shall we

11
00:00:49,840 --> 00:00:58,399
now nvidia doesn't actually manufacture display panels that work is best left to

12
00:00:55,360 --> 00:01:00,320
the experts at samsung lg display

13
00:00:58,399 --> 00:01:04,960
au optronics and the like but inside this cage come on get around

14
00:01:03,039 --> 00:01:12,799
here behind me is a sample of every g-sync monitor that

15
00:01:09,439 --> 00:01:15,840
has ever been built and the thing is

16
00:01:12,799 --> 00:01:18,479
there's far more to creating a great

17
00:01:15,840 --> 00:01:22,640
finished gaming display than just taking whatever panel those guys hand you

18
00:01:20,720 --> 00:01:26,200
slapping it into an enclosure and calling it a day

19
00:01:27,119 --> 00:01:36,560
in fact while the situation has improved a lot especially on the desktop

20
00:01:32,320 --> 00:01:39,040
nvidia still fails over 50 percent

21
00:01:36,560 --> 00:01:42,960
of the laptop displays that get sent to them for validation

22
00:01:40,880 --> 00:01:46,399
so each one of these assuming that they don't immediately

23
00:01:44,400 --> 00:01:52,399
fail the more basic checks will get subjected to over

24
00:01:48,360 --> 00:01:54,479
300 tests some of which take hours or

25
00:01:52,399 --> 00:01:57,600
even multiple days to run

26
00:01:55,520 --> 00:02:02,479
so i feel kind of bad for the techs who drew the short straw for our visit today

27
00:01:59,759 --> 00:02:06,479
and had to shut down you see the issue and actually the reason that the walls

28
00:02:04,240 --> 00:02:11,440
are all painted black in here is that with our filming lights on they can't

29
00:02:08,640 --> 00:02:16,319
run any real tests because the results will be invalid so we're gonna try to

30
00:02:13,599 --> 00:02:21,520
get out of their way real quick here on this bench then is our first real

31
00:02:18,879 --> 00:02:24,760
test here so it's a pretty simple one on the surface

32
00:02:22,720 --> 00:02:30,080
but it's one that will weed out a surprising number of variable refresh

33
00:02:27,040 --> 00:02:32,640
rate displays so our gpu here runs a

34
00:02:30,080 --> 00:02:38,400
benchmark with extreme swings in the frame rate which is designed to expose

35
00:02:35,040 --> 00:02:40,160
weaknesses in the t-con or timing

36
00:02:38,400 --> 00:02:44,400
controller which is the part of the panel that takes the output from the

37
00:02:42,000 --> 00:02:49,920
scalar and translates it into something that the actual drivers of the tft

38
00:02:47,360 --> 00:02:54,879
matrix can understand now the results of these failures are usually quite obvious

39
00:02:52,640 --> 00:03:00,640
because if a t-con experiences a firmware problem for example causing an

40
00:02:57,280 --> 00:03:03,760
overflow condition its go-to emergency

41
00:03:00,640 --> 00:03:06,319
response is generally to blank out parts

42
00:03:03,760 --> 00:03:10,319
of the panel or even the entire thing temporarily to prevent it from being

43
00:03:08,239 --> 00:03:13,440
damaged apparently they actually had one of those come through here just the

44
00:03:11,599 --> 00:03:19,200
other day but that doesn't mean that running these

45
00:03:15,760 --> 00:03:21,680
tests is simple for one thing these

46
00:03:19,200 --> 00:03:25,920
kinds of problems can be weird edge cases i mean that's why the manufacturer

47
00:03:23,920 --> 00:03:32,000
didn't notice them so they can take a long time to manifest which is why they

48
00:03:28,480 --> 00:03:33,120
run for up to 48 hours before assigning

49
00:03:32,000 --> 00:03:39,120
a pass and for another just hooking a source up to a bare

50
00:03:37,519 --> 00:03:44,640
notebook lcd is an ordeal and a half

51
00:03:41,120 --> 00:03:47,440
so of course you need a mobile gpu in

52
00:03:44,640 --> 00:03:51,519
order to validate mobile displays but using a standard desktop test bench

53
00:03:49,599 --> 00:03:56,720
would save you a lot of time so check out this frankenstein creation i've

54
00:03:55,120 --> 00:04:03,680
actually got another one right here so this according to the folks here

55
00:03:59,519 --> 00:04:06,959
is a touring mobile gpu but soldered

56
00:04:03,680 --> 00:04:08,159
onto a development board with just like

57
00:04:06,959 --> 00:04:12,239
this cooler just plugs straight into 12 volt power

58
00:04:10,319 --> 00:04:17,440
it just runs at full speed all the time that's why these things are so loud and

59
00:04:14,080 --> 00:04:20,959
then it's just bristling with diagnostic

60
00:04:17,440 --> 00:04:22,639
readouts and and uh sort of measurement

61
00:04:20,959 --> 00:04:26,320
points and then there's then there's the really cool stuff so this right here

62
00:04:25,360 --> 00:04:32,720
is our dvi output or something that's on a

63
00:04:30,080 --> 00:04:38,000
daughter board for some reason and on the back here this is really cool

64
00:04:35,520 --> 00:04:43,199
so these interfaces back here are designed for all of the different

65
00:04:40,160 --> 00:04:45,520
implementations of embedded displayport

66
00:04:43,199 --> 00:04:50,000
that you might see from the various panel manufacturers

67
00:04:47,440 --> 00:04:56,320
so they make their own daughterboard pcbs here so they can adapt this weird

68
00:04:53,440 --> 00:05:00,960
development card to run any display that they want on to our next test though one

69
00:04:59,199 --> 00:05:05,680
of the hardest parts of creating a variable refresh rate display is

70
00:05:03,040 --> 00:05:10,400
preventing flickering but the thing is not everyone perceives flicker the same

71
00:05:08,560 --> 00:05:15,199
and even if a flicker isn't visible to the eye it can still cause fatigue and

72
00:05:12,720 --> 00:05:19,039
headaches so that is where this test comes in

73
00:05:16,160 --> 00:05:24,160
now it doesn't look like much but the lab techs here can change this boring

74
00:05:21,440 --> 00:05:29,280
gray screen to output any refresh rate they want then they use this klein

75
00:05:26,720 --> 00:05:35,120
instrument's k10a a basic luminance meter to measure the amplitude and

76
00:05:31,759 --> 00:05:37,840
decibels of any changes in the output

77
00:05:35,120 --> 00:05:41,680
brightness then that is of course assuming that the panel makes it this

78
00:05:39,360 --> 00:05:47,600
far they use this special box of nvidia's own creation to

79
00:05:45,520 --> 00:05:52,800
kind of like a like a doctor checking your like heartbeat or something check

80
00:05:50,320 --> 00:05:58,320
all the different areas of the panel to see which one is the worst and then

81
00:05:55,440 --> 00:06:03,520
ensure that even that is still within spec rounding out panel selection is a

82
00:06:01,039 --> 00:06:08,400
whole battery of other tests color reproduction color gamut absolute

83
00:06:05,840 --> 00:06:14,560
luminance native contrast ratio pixel response times you pretty much name it

84
00:06:11,680 --> 00:06:18,560
then we can move into this room once a panel is validated it can move on to

85
00:06:17,120 --> 00:06:25,039
stage two actual display development at this point

86
00:06:21,919 --> 00:06:27,600
things get a lot pickier so this forty

87
00:06:25,039 --> 00:06:32,960
to fifty thousand dollar xy positioning jib is precise to one tenth of a

88
00:06:30,000 --> 00:06:37,120
millimeter so what they do is they light up the display so the one that we're

89
00:06:34,880 --> 00:06:42,400
looking at right here is a g-sync ultimate unit with hdr 10 support at up

90
00:06:40,000 --> 00:06:47,680
to 1 000 nits peak brightness and then they take measurements across the entire

91
00:06:45,039 --> 00:06:53,520
surface to ensure that it's uniform it's pretty freaking intense except

92
00:06:50,319 --> 00:06:55,880
that is just scratching the surface so

93
00:06:53,520 --> 00:07:00,720
this display also features 384 zone full array backlighting and

94
00:06:59,360 --> 00:07:07,599
handling that evenly is really freaking tricky in this test

95
00:07:05,199 --> 00:07:12,080
nvidia is evaluating each of the individual leds that makes up the

96
00:07:09,680 --> 00:07:17,840
backlight array not just off and at full force but at its various

97
00:07:15,280 --> 00:07:23,280
steps in between because you guys gotta understand they need to account not just

98
00:07:20,880 --> 00:07:28,639
for the drive level of each one of these zones but even the neighboring light

99
00:07:26,000 --> 00:07:34,000
lead from other zones around them and then making matters worse the whole

100
00:07:31,199 --> 00:07:39,759
thing has got to work perfectly at every brightness level and every refresh rate

101
00:07:38,240 --> 00:07:44,479
so then once the luminance behavior of the panel

102
00:07:41,680 --> 00:07:48,000
is characterized it's either fantastic out of the box ready to go

103
00:07:46,880 --> 00:07:54,319
or much more likely it needs some work it's

104
00:07:50,880 --> 00:07:56,319
no bloody wonder that these 4k 144hz hdr

105
00:07:54,319 --> 00:08:01,360
displays got delayed right speaking of delays in this room

106
00:07:59,440 --> 00:08:05,199
also uh professionally darkened out don't you love it when like

107
00:08:03,120 --> 00:08:09,520
an engineer is given a problem and told to solve it quickly i love this um

108
00:08:08,000 --> 00:08:14,319
anyway in here is one of nvidia's bfgd or big format

109
00:08:12,720 --> 00:08:18,319
gaming displays now the process you're seeing here is

110
00:08:17,039 --> 00:08:26,560
nothing special if you've ever professionally calibrated a display before basically a signal generator so

111
00:08:23,759 --> 00:08:31,280
this laptop outputs a known value let's say white or other more different white

112
00:08:28,639 --> 00:08:36,560
or red or something and then a sensor like this spectroradiometer checks how

113
00:08:34,000 --> 00:08:42,320
close it is to what it's expecting but what's different about nvidia's approach

114
00:08:39,039 --> 00:08:44,080
here is that instead of just calibrating

115
00:08:42,320 --> 00:08:48,240
finished displays during the development process they're

116
00:08:46,320 --> 00:08:53,200
jumping in to make sure that there are no underlying issues with the technology

117
00:08:51,279 --> 00:08:58,560
that are going to crop up while operating in variable refresh rate mode

118
00:08:56,080 --> 00:09:02,480
that can't be corrected because the thing is

119
00:08:59,600 --> 00:09:08,000
many lcd characteristics change when the refresh rate changes another

120
00:09:05,200 --> 00:09:11,680
fun one to deal with is overdrive so overdrive basically works like this

121
00:09:10,160 --> 00:09:17,440
you've got a pixel and you want to take it from

122
00:09:13,760 --> 00:09:20,399
level 100 to level 200 which might take

123
00:09:17,440 --> 00:09:25,120
it let's say 8 milliseconds natively well if that's not good enough for the

124
00:09:22,399 --> 00:09:30,800
performance you want with overdrive we can tell that pixel i want 250. and

125
00:09:28,560 --> 00:09:36,240
because that's a bigger change it's going to reach our actual desired level

126
00:09:33,360 --> 00:09:39,760
of 200 faster so you're basically giving it a bigger kick so you might get there

127
00:09:38,560 --> 00:09:45,040
in just four or six milliseconds

128
00:09:41,839 --> 00:09:47,120
the trick though is to not go too high

129
00:09:45,040 --> 00:09:50,800
or you'll get too high of a value overshooting the intended target

130
00:09:49,440 --> 00:09:56,959
now here's the thing at 60 hertz fixed refresh rate

131
00:09:54,080 --> 00:10:01,920
this is relatively simple so the monitor just has a lookup table built into it of

132
00:10:00,080 --> 00:10:07,279
what it should shoot for in order to have the value be correct by the next

133
00:10:04,120 --> 00:10:10,640
16.6 millisecond refresh cycle

134
00:10:07,279 --> 00:10:13,760
with variable refresh rate well it has

135
00:10:10,640 --> 00:10:16,399
to compensate how much of an extra kick

136
00:10:13,760 --> 00:10:22,880
it's giving according to how much more or how much less time it's going to have

137
00:10:18,560 --> 00:10:24,880
to get to the actual desired value

138
00:10:22,880 --> 00:10:29,519
and even driving the leds in the backlight is a complicated matter so

139
00:10:27,120 --> 00:10:34,000
this is the back of a g-sync ultimate desktop monitor but the internal guts

140
00:10:32,000 --> 00:10:37,680
here are actually fundamentally similar to the bfgd that we just saw so at its

141
00:10:36,800 --> 00:10:43,440
heart is the powerful g-sync processor that

142
00:10:39,920 --> 00:10:46,560
contains much of g-sync's special sauce

143
00:10:43,440 --> 00:10:49,839
then around it is kind of like the uh

144
00:10:46,560 --> 00:10:51,680
the motherboard of a monitor so

145
00:10:49,839 --> 00:10:56,240
this here is something that the manufacturer of the monitor would

146
00:10:53,839 --> 00:11:03,200
customize depending on what they want for display connectors usb hubs

147
00:11:00,399 --> 00:11:07,200
built-in audio that sort of thing then finally there's a third part here so

148
00:11:06,240 --> 00:11:14,640
these are kind of like the wings that make g-sync hdr fly

149
00:11:11,600 --> 00:11:17,279
so these driver ics right here take a

150
00:11:14,640 --> 00:11:21,880
digital signal from the g-sync module scaler for what the luminance level

151
00:11:19,519 --> 00:11:29,279
should be for each of the 384 leds that make up the backlight here

152
00:11:25,440 --> 00:11:32,640
then they output a given dc voltage to

153
00:11:29,279 --> 00:11:35,360
each of the 16 transistors that drive

154
00:11:32,640 --> 00:11:41,120
the current that actually lights up the individual led in each zone

155
00:11:38,320 --> 00:11:44,880
just the right amount but like hold on hold on a second here

156
00:11:42,640 --> 00:11:52,000
a corsair commander pro can drive dozens of rgb leds per channel

157
00:11:49,360 --> 00:11:58,720
with like four wires what is all of this for well here's the

158
00:11:55,279 --> 00:12:01,440
thing at 144 hertz you've got under

159
00:11:58,720 --> 00:12:07,200
seven milliseconds per frame and you have to both determine and write all 384

160
00:12:05,600 --> 00:12:12,240
of those values each frame and add up to a thousand

161
00:12:10,079 --> 00:12:16,639
nitspeak brightness so this is not a trivial task like man i can imagine the

162
00:12:14,959 --> 00:12:19,040
meeting where they decided to build this thing right now

163
00:12:18,000 --> 00:12:25,279
like all right team so we've never built a tv before

164
00:12:23,440 --> 00:12:31,120
so the plan is to build the most difficult one

165
00:12:26,560 --> 00:12:31,120
lol good luck everyone and break

166
00:12:31,200 --> 00:12:38,720
which leads us finally to certification now the third and final stage might not

167
00:12:36,399 --> 00:12:43,200
be that visually interesting but it's arguably the most important so this is

168
00:12:41,279 --> 00:12:48,560
the point at which nvidia receives the first finished units of each display and

169
00:12:46,240 --> 00:12:54,160
goes through that whole ordeal again to ensure that nothing got lost in

170
00:12:51,519 --> 00:12:59,519
translation and i have to say going behind the scenes today gave me a

171
00:12:56,560 --> 00:13:04,079
much deeper appreciation for what nvidia has been doing in the display industry

172
00:13:02,079 --> 00:13:08,720
the eagle-eyed among you for example might have noticed that one of those

173
00:13:06,800 --> 00:13:14,480
backlight driver boards on this prototype has an nvidia silkscreen on it

174
00:13:12,079 --> 00:13:18,240
while the other one has something else

175
00:13:16,000 --> 00:13:24,000
that's because as part of developing g-sync ultimate nvidia actually created

176
00:13:21,920 --> 00:13:30,560
the reference design for these driver boards since nobody had ever done a

177
00:13:27,279 --> 00:13:32,800
384-zone 27-inch panel before now

178
00:13:30,560 --> 00:13:37,680
unfortunately i wasn't able to show you guys everything that i saw but what i

179
00:13:35,600 --> 00:13:42,720
can say is that my experience today has shown me definitively that this is far

180
00:13:40,800 --> 00:13:47,120
more than just a a rubber stamping

181
00:13:44,800 --> 00:13:53,200
certification fee operation so then there are three tiers of g-sync

182
00:13:49,680 --> 00:13:55,360
now g-sync compatible displays don't go

183
00:13:53,200 --> 00:13:59,519
through anything that you saw today nvidia performs four variable refresh

184
00:13:57,920 --> 00:14:04,800
rate tests to ensure that they're suitable for a basic vrr gaming

185
00:14:01,920 --> 00:14:09,120
experience then there's g-sync and g-sync ultimate where you're getting the

186
00:14:07,120 --> 00:14:14,320
deep collaboration between your graphics card manufacturer and your display maker

187
00:14:12,000 --> 00:14:21,120
with ultimate also including support for hdr gaming which as we've discussed in

188
00:14:16,639 --> 00:14:23,680
the past looks pretty freaking sick

189
00:14:21,120 --> 00:14:27,760
so that's it for today guys massive thank you to you for watching and to

190
00:14:25,680 --> 00:14:33,120
nvidia for sponsoring this video and just straight up allowing us a peek

191
00:14:31,040 --> 00:14:37,519
behind the kimono this was absolutely incredible if this video sucked you guys

192
00:14:35,680 --> 00:14:40,880
know what to do but if it was awesome get subscribed hit the like button or

193
00:14:39,600 --> 00:14:45,839
check out the link to where to buy the stuff we featured in the video description also down there is our merch

194
00:14:44,320 --> 00:14:49,680
store which has cool shirts like this one and our community forum which you

195
00:14:48,160 --> 00:14:53,360
should totally join especially if you have any questions about building a

196
00:14:51,279 --> 00:14:57,839
fantastic gaming rig or buying a variable refresh rate monitor or

197
00:14:54,639 --> 00:14:57,839
whatever the case may be