1
00:00:00,000 --> 00:00:08,400
I want you to stop typing on garbage and

2
00:00:06,299 --> 00:00:12,960
I've got just the thing to make it happen with some help from our Labs team

3
00:00:10,380 --> 00:00:18,240
and my new robot friend here I intend to make sure that you never get fooled by

4
00:00:15,240 --> 00:00:25,439
flashy marketing again it presses it stresses it undresses hey yeah not all

5
00:00:22,500 --> 00:00:29,820
the cameras on you yep cave here is programmed to do all sorts of useful

6
00:00:27,359 --> 00:00:35,100
things but most importantly he's designed to give you some of the most

7
00:00:31,980 --> 00:00:38,219
accurate and precise keyboard data on

8
00:00:35,100 --> 00:00:40,559
the market so no manufacturer can get

9
00:00:38,219 --> 00:00:44,520
away with BS in their product specs ever again be afraid keyboard makers be very

10
00:00:43,559 --> 00:00:49,739
afraid afraid of this segue to our sponsor

11
00:00:46,879 --> 00:00:54,600
zoho1 Zoho one brings together sales marketing Finance analytics and support

12
00:00:52,079 --> 00:00:59,219
in a single unified platform to run your entire business try Zoho won today for

13
00:00:57,000 --> 00:01:04,080
30 days using the link down below when your Corsair k70 Pro or Logitech g915

14
00:01:02,280 --> 00:01:08,400
rolls off the production line it goes through a series of QC or quality

15
00:01:06,180 --> 00:01:11,820
control checks to make sure that every little piece of it is within the

16
00:01:09,840 --> 00:01:16,020
manufacturer's specified tolerances do the switches work do they chatter does

17
00:01:13,860 --> 00:01:19,860
the RGB light does the chassis line up is the cable freighter the logos upside

18
00:01:17,939 --> 00:01:23,880
down or off center all of these things and more are checked before your

19
00:01:21,960 --> 00:01:29,520
keyboard goes into the box out the door and finally lands on your desk but a

20
00:01:27,299 --> 00:01:35,700
quality controlled product is not necessarily a quality product Brands

21
00:01:33,000 --> 00:01:40,799
love to use words like precise but then stay conveniently quiet when you ask for

22
00:01:38,040 --> 00:01:45,720
details how precise what is your specified error tolerance I mean think

23
00:01:42,600 --> 00:01:47,939
about it if the force to push down W and

24
00:01:45,720 --> 00:01:52,979
S let's say was allowed to be plus or minus ten percent the difference could

25
00:01:49,979 --> 00:01:54,780
be as much as 20 between them is that

26
00:01:52,979 --> 00:01:58,380
going to cost you a critical hit what if you were to push the space bar here

27
00:01:56,159 --> 00:02:03,659
instead of here will it feel the same will it work at all with the flex of

28
00:02:00,600 --> 00:02:05,399
ryzen 4S we intend to find out it's got

29
00:02:03,659 --> 00:02:10,679
seven degrees of freedom can move payloads as heavy as three and a half

30
00:02:07,140 --> 00:02:13,140
kilograms and can reach over 780

31
00:02:10,679 --> 00:02:17,700
millimeters we've spent months getting this guy up and running to a degree that

32
00:02:15,000 --> 00:02:22,500
we deem satisfactory to start sharing the data with you okay second Jake show

33
00:02:20,160 --> 00:02:27,180
me what you got is Cherry mx's plus minus 10 accurate what what actually

34
00:02:25,200 --> 00:02:31,260
works so far well we haven't actually started Cherry MX but on this particular

35
00:02:29,700 --> 00:02:34,440
keyword what we've done is we've taken it we've placed it into the build area

36
00:02:32,700 --> 00:02:38,520
and tightened these little screws up against it so that it doesn't move it is

37
00:02:36,599 --> 00:02:40,920
moving that's well only if you push it the robot's not going to move it what

38
00:02:39,840 --> 00:02:45,480
we're doing is we're taking the computer vision okay and we're actually taking a

39
00:02:43,920 --> 00:02:49,019
series of different frames we're composing those over top of each other

40
00:02:46,920 --> 00:02:52,440
after we do Edge detection so we can actually see where each of the keys are

41
00:02:50,760 --> 00:02:57,180
so we can tell the size of the key the shape of the key so once that's done the

42
00:02:55,319 --> 00:03:01,500
robot will move down onto the center point of each key press down measure the

43
00:02:59,760 --> 00:03:04,800
force as it travels and then measure it as it releases as well doing this gives

44
00:03:03,480 --> 00:03:07,800
us not only the force graph for the switches being used on the board but

45
00:03:06,480 --> 00:03:13,260
also it lets us know if any of the switches are out of spec but how

46
00:03:09,660 --> 00:03:15,000
accurate slash precises it what is this

47
00:03:13,260 --> 00:03:19,860
exactly it looks a little suggestive this is Antoine's handiwork it wasn't my

48
00:03:17,340 --> 00:03:24,840
first choice okay what this is this is a 3D printed finger and effectively it's a

49
00:03:22,980 --> 00:03:28,800
finger yeah it's a finger for sure PG show this is actually connected to the

50
00:03:26,519 --> 00:03:34,080
force load cell built into the robot so when we press down we're actually able

51
00:03:30,480 --> 00:03:37,620
to record the Precision at 0.001 Newton

52
00:03:34,080 --> 00:03:39,180
so really really fine detail this is way

53
00:03:37,620 --> 00:03:43,260
better than what we had during our like last tour where we had like a prototype

54
00:03:40,799 --> 00:03:46,019
thing of this oh way way better yeah we replaced the previous system because

55
00:03:44,340 --> 00:03:51,180
because the Gantry system that we had actually has issues with moving and

56
00:03:48,840 --> 00:03:54,060
vibration this is really stable with a high amount of repeatability which means

57
00:03:52,920 --> 00:03:58,620
that we're not going to run into those issues and dirty our data we want things

58
00:03:56,519 --> 00:04:02,280
to be cleanliness is it supposed to be doing things like were we going to start

59
00:04:00,299 --> 00:04:05,400
it at some point yeah as soon as as soon as we have a cue Antoine can hit e and

60
00:04:04,080 --> 00:04:10,680
it'll start moving yeah yeah yeah yeah it will move fast whoa hey whoa how you

61
00:04:08,220 --> 00:04:15,000
doing okay well yeah do the thing I mean did you stop it no he didn't but we

62
00:04:12,959 --> 00:04:18,540
moved to the keyboard that's all is this one of those things where you like get

63
00:04:16,260 --> 00:04:21,479
the demo ready for the CEO and then the CEO leaves and it didn't actually it

64
00:04:20,100 --> 00:04:27,300
wasn't actually working of course this is all pre-programmed and then you move the keyboard it took an image of the

65
00:04:25,259 --> 00:04:31,199
area and detected where the keys were yeah and then we moved to the keyboard

66
00:04:29,100 --> 00:04:34,139
yeah I see so obviously it hasn't can't see it okay it's too close to a new

67
00:04:32,580 --> 00:04:38,160
image yeah we can do it I guess if I'm like bumping this in the middle of the thing that ruins everything oh yeah

68
00:04:36,720 --> 00:04:42,479
we're not going to use this data for any publishing this is just to show you and

69
00:04:40,740 --> 00:04:46,919
to show liners what you're seeing here is an HSV version of the camera image

70
00:04:45,120 --> 00:04:50,400
and this is for calibration purposes so if the camera gets bumped or moves what

71
00:04:49,320 --> 00:04:55,380
we're going to do is we're going to find these four points here and we're going to do a d skew on the image to make sure

72
00:04:53,759 --> 00:04:59,340
that we're looking at a level field so once we've done that it's going to take

73
00:04:57,000 --> 00:05:02,460
the 50 frames of video run the edge detection on each of them and then

74
00:05:00,720 --> 00:05:06,660
composite those together and that way if there's any autofocus issues we're able

75
00:05:04,500 --> 00:05:11,820
to bypass that how long does that take it's done right now it's not 100 perfect

76
00:05:09,300 --> 00:05:16,199
but we have a new computer vision higher who it's her specialty to make things

77
00:05:14,340 --> 00:05:19,979
like this work we're working on it and actually we're looking at moving away

78
00:05:17,699 --> 00:05:23,940
from computer vision to a 3D laser profiler and that we hired a person to

79
00:05:22,259 --> 00:05:28,139
do this and then immediately they're out of a job no no no good she's actually

80
00:05:26,520 --> 00:05:31,800
she's a machine learning and computer vision specialist the same skill set

81
00:05:30,180 --> 00:05:36,199
applies to the 3D image that we're gonna get okay from the profiler okay but it's

82
00:05:34,320 --> 00:05:40,860
laser thing it's a laser with lasers okay yeah no sharks but lasers okay I'm

83
00:05:39,479 --> 00:05:44,580
just gonna ask whatever it's attached to can we name it shark what the 3D laser

84
00:05:43,440 --> 00:05:49,320
profile is going to do is really clean up the image we don't have to worry about lighting we don't have to worry

85
00:05:46,860 --> 00:05:52,860
about changing the environment and because it's just looking at the height

86
00:05:50,940 --> 00:05:57,479
map all of the gaps between each of the keys presents a huge distinct drop right

87
00:05:56,100 --> 00:06:02,460
meaning that we're going to get every key every time whereas right now we get

88
00:05:59,940 --> 00:06:08,400
them Shadows we've got I guess Ah that's the point of this this is the light

89
00:06:05,160 --> 00:06:09,900
blocker 9000 and they did not want to

90
00:06:08,400 --> 00:06:15,600
have this on here for the shoot because it looks janky but what we had to do was

91
00:06:13,080 --> 00:06:19,800
put this on here to get even this good of an image because once you change your

92
00:06:17,940 --> 00:06:24,479
lighting and have you know a shadow like this it's really easy to get an extra

93
00:06:22,319 --> 00:06:27,960
key detected on that line what we do have is the ability for the person

94
00:06:26,039 --> 00:06:32,100
running the test to actually remove and add boxes if they need to after a full

95
00:06:30,060 --> 00:06:36,479
test run of 100 some odd switches cave is going to tell us with certainty how

96
00:06:34,199 --> 00:06:41,160
many of them actuate or send a signal to your PC whether they actuate early or

97
00:06:39,120 --> 00:06:45,419
late compared to the spec whether the travel distance being reported by the

98
00:06:42,720 --> 00:06:50,039
manufacturer is accurate if the bump in a tactile switch is where they say it is

99
00:06:47,400 --> 00:06:54,419
and these might sound like very boring little details to you but they're the

100
00:06:52,139 --> 00:06:59,039
kind of thing that you don't realize is a problem until you are thighs deep in a

101
00:06:57,240 --> 00:07:04,199
gaming or writing session and you keep accidentally freaking crumpling that

102
00:07:01,919 --> 00:07:08,400
line any error from our point of view should be the fault of the user and not

103
00:07:06,120 --> 00:07:13,259
of the peripheral and small deviations like this can make a huge difference

104
00:07:10,680 --> 00:07:17,039
when muscle memory is at play do you guys remember this viral video showing

105
00:07:14,880 --> 00:07:21,780
people tripping on 36th Street subway station in New York that stare was less

106
00:07:19,500 --> 00:07:26,759
than an inch out of spec that's less than 15 percent of course this only

107
00:07:24,060 --> 00:07:31,740
helps us if we can test every keyboard can we do that oh buddy let me tell you

108
00:07:29,580 --> 00:07:36,840
right now we're able to test this keyboard in 30 minutes and I know a few

109
00:07:34,979 --> 00:07:40,139
places we can actually find Time Savings I think we can get it down to 25. I can

110
00:07:38,940 --> 00:07:43,979
see one place we could probably save some time like if we have that laser

111
00:07:42,000 --> 00:07:48,960
depth map we probably wouldn't need to start like exactly the key and then go

112
00:07:46,560 --> 00:07:53,639
down slowly and I guess the slowness is so that we get nice clean readings

113
00:07:51,360 --> 00:07:58,199
exactly it so right now we're actually going to a fixed height that is higher

114
00:07:56,220 --> 00:08:01,259
than any known keyboard we've found right okay and that way we know we're

115
00:07:59,819 --> 00:08:04,620
never going to crash into it because that would be bad it would be very bad

116
00:08:02,940 --> 00:08:08,699
like it's an expensive robot you'd go right through the keyboard and probably

117
00:08:06,479 --> 00:08:13,560
damage the robot tell me something though no user would ever press a key

118
00:08:10,740 --> 00:08:17,759
like that so so after we hit everything slowly have we considered going back and

119
00:08:15,780 --> 00:08:21,060
doing a fast hit and looking at what that looks like we have we've also

120
00:08:19,259 --> 00:08:23,759
actually considered hitting the corners as well for checking for key binding

121
00:08:22,440 --> 00:08:28,379
I've just been told that our new procurement hire has 150 keyboards on

122
00:08:26,340 --> 00:08:32,039
order already with our goal being to test 200 by the end of this year

123
00:08:30,180 --> 00:08:36,120
starting with the big names like Logitech Corsair and Razer then moving

124
00:08:34,380 --> 00:08:41,459
on to the other top selling keyboards on Amazon but that only tells us about the

125
00:08:38,640 --> 00:08:45,300
field what about latency or other concerns like repairability what this

126
00:08:44,039 --> 00:08:51,240
history does is measuring the total system latency so

127
00:08:49,200 --> 00:08:55,680
from the time you click on the keyboard to the time you have a response up on

128
00:08:53,940 --> 00:08:59,220
the display like Amazon flash to automate this process we've got a

129
00:08:57,360 --> 00:09:04,680
solenoid right here that presses the key for us then once the key is pressed the

130
00:09:02,160 --> 00:09:09,839
signal goes to two places it goes to this NVIDIA ldat detector box and it

131
00:09:07,680 --> 00:09:14,339
goes to the accompanying software on the PC by measuring the difference in time

132
00:09:12,120 --> 00:09:18,720
between the signal reaching here and the muzzle flash appearing here we can find

133
00:09:16,440 --> 00:09:23,160
out how long it takes between pressing a key on your keyboard and your character

134
00:09:20,459 --> 00:09:27,320
moving on screen this kind of real world testing is really important because it

135
00:09:25,200 --> 00:09:32,399
allows us to quantify the importance of say high polling rates in keyboards

136
00:09:30,300 --> 00:09:37,019
it'll also help us discover if there's more to latency than just the pulling

137
00:09:34,740 --> 00:09:40,320
rate it's pretty rudimentary for now and what we want to do in the longer term is

138
00:09:38,640 --> 00:09:45,240
set up a USB breakout circuit that's tied to a logic analyzer then watch the

139
00:09:42,959 --> 00:09:48,000
USB packet as it crosses the wire but that's going to take a little bit of

140
00:09:46,560 --> 00:09:51,660
time for those of you wondering by the way for Bluetooth we'll just be using a

141
00:09:49,680 --> 00:09:55,620
dongle one question I have about this is right now we're using a 360 Hertz

142
00:09:53,580 --> 00:10:01,860
monitor which is the fastest available for now but as faster monitors become

143
00:09:59,040 --> 00:10:06,600
available will that make our old data no longer applicable if we switch and use

144
00:10:03,899 --> 00:10:11,100
say a 500 Hertz monitor dear data will still be relevant because my goal is to

145
00:10:08,820 --> 00:10:16,500
have the end-to-end latency and still subtracting the system so just the PC

146
00:10:14,580 --> 00:10:20,640
and the display from the keyboard latency so you would have just a

147
00:10:18,600 --> 00:10:24,600
keyboard latency okay so this part would still make sense in the long run right

148
00:10:22,500 --> 00:10:28,700
how are we going to remove the display latency from the keyboard latency

149
00:10:26,640 --> 00:10:34,800
locally NVIDIA in the ldat system give us the the PC

150
00:10:33,300 --> 00:10:40,080
latency so we just take that value subtract it oh this is being said I'm

151
00:10:38,040 --> 00:10:45,980
not a big fan of that method that's why we are working on on the USB breaker and

152
00:10:42,839 --> 00:10:49,140
Music Analyzer just

153
00:10:45,980 --> 00:10:51,300
the idea is to have both value or real

154
00:10:49,140 --> 00:10:56,160
world value sure and the theoretical value and the reason that that matters

155
00:10:53,459 --> 00:11:01,019
is because you might see a theoretical value that's half the time of another

156
00:10:58,740 --> 00:11:04,800
keyboard but in terms of the difference that it's going to make to an actual

157
00:11:02,640 --> 00:11:08,519
game might be functionally zero because that makes up such a small portion of

158
00:11:07,019 --> 00:11:13,440
the total latency that it's just negligible for example you can have a

159
00:11:10,440 --> 00:11:15,480
keyboard that is one millisecond

160
00:11:13,440 --> 00:11:20,220
um as the late 107 milliseconds or two milliseconds seems like a lot like two

161
00:11:17,760 --> 00:11:23,339
times more but when you take the full end-to-end latency it might be the

162
00:11:21,720 --> 00:11:27,420
difference between 15 and 60 milliseconds how the hell are you going

163
00:11:25,200 --> 00:11:31,980
to test 200 keyboards by the end of this that's a great question

164
00:11:29,420 --> 00:11:36,300
that's a wonderful question because it's going to be you right so far just me of

165
00:11:34,800 --> 00:11:40,380
course there's a lot more to evaluating a keyboard than just collecting data

166
00:11:38,459 --> 00:11:44,459
about the key switches and the latency one of the other big concerns in the lab

167
00:11:42,060 --> 00:11:48,120
is repairability for instance how easy is it to access the screws do you have

168
00:11:46,320 --> 00:11:51,720
to pull off little rubber feet that you then can't read here once we have our

169
00:11:50,160 --> 00:11:55,800
website up and running more on that later we aim to list all the screw types

170
00:11:53,940 --> 00:11:58,860
and tools required whether or not there are plastic Clips holding pieces

171
00:11:57,060 --> 00:12:02,579
together any extra steps you might need to get it apart and tear down pictures

172
00:12:00,839 --> 00:12:06,000
so you can see for yourself what's inside the board you're buying we'll

173
00:12:04,140 --> 00:12:09,540
also be giving points for whether or not boards have hot swappable pcbs for

174
00:12:07,800 --> 00:12:13,980
switch Replacements and detachable cables because not having to solder in

175
00:12:12,240 --> 00:12:17,940
order to repair your keyboard makes it much easier for regular users to keep

176
00:12:15,959 --> 00:12:21,480
their Electronics out of the landfill we also want to use our articles to

177
00:12:19,560 --> 00:12:26,459
contextualize this information I mean it's all fine and good to see a key

178
00:12:23,399 --> 00:12:29,160
switch actuation Force graph but how

179
00:12:26,459 --> 00:12:33,060
does it feel our aim is to include comparisons then against popular

180
00:12:31,019 --> 00:12:37,380
switches and keyboards to help set you in the right direction do you like the

181
00:12:35,220 --> 00:12:41,700
feel of MX Blues but want something quieter do you like the actuation force

182
00:12:39,600 --> 00:12:45,720
of your Amazon Basics keyboard but wish it felt less mushy and looked nicer

183
00:12:44,040 --> 00:12:48,839
we've still got a long way to go we already talked about moving to a

184
00:12:47,100 --> 00:12:53,279
high-speed 3D laser profiler system instead of a camera so that we can get

185
00:12:50,820 --> 00:12:57,300
3D images of each board and the website isn't anywhere close to being ready

186
00:12:55,019 --> 00:13:00,959
while we do have several test plans drafted there are still a lot of

187
00:12:58,920 --> 00:13:05,160
questions to answer around ergonomics and dimensions and key rollover and

188
00:13:03,240 --> 00:13:09,060
ghosting connectivity bundle and packaging software firmware backlighting

189
00:13:07,079 --> 00:13:13,200
extra controls and of course sound these are all things we haven't

190
00:13:11,399 --> 00:13:17,100
even touched on today but we do have plans in the works for them so stay

191
00:13:14,820 --> 00:13:21,120
tuned the ultimate goal is to create an information repository with a detailed

192
00:13:19,200 --> 00:13:25,740
breakdown of test results pictures from every angle maybe a rubric scoring

193
00:13:23,880 --> 00:13:29,700
system I'll be arguing hard against it but it might be unavoidable and most

194
00:13:28,139 --> 00:13:33,180
importantly to all the enthusiasts out there a sound Library where you can

195
00:13:31,500 --> 00:13:36,360
listen to each keyboard in its stock configuration one of the biggest

196
00:13:34,620 --> 00:13:40,380
problems with keyboard audio tests online is a lack of consistency which we

197
00:13:38,579 --> 00:13:45,779
hope to solve by using the same space and desk materials namely deskpad from

198
00:13:43,440 --> 00:13:49,740
lttstore.com as well as the same microphone so that you get the closest

199
00:13:47,820 --> 00:13:54,600
sound possible to hearing it in person lofty goals indeed and as you can see

200
00:13:52,320 --> 00:14:00,300
we're not there yet but you know who is there our sponsor our sponsor is always

201
00:13:57,120 --> 00:14:02,639
there ASUS ASUS's long-awaited GeForce

202
00:14:00,300 --> 00:14:06,720
RTX 4000 series graphics cards are finally here so you'll be able to get

203
00:14:04,139 --> 00:14:12,959
your hands on their top tier GeForce RTX 4090 4080 16 gig and 140 80 12 gig these

204
00:14:10,800 --> 00:14:17,519
cards boast plenty of Cuda cores heaps of video memory and have all the GeForce

205
00:14:15,120 --> 00:14:21,899
RTX features to make sure your build is the best of the best there are 40 90 Rog

206
00:14:19,920 --> 00:14:25,800
strix offers a new patented Vapor chamber with a milled heat spreader for

207
00:14:23,339 --> 00:14:30,060
lower GPU temps and a massive fin array that's optimized for airflow from three

208
00:14:27,480 --> 00:14:33,899
axial tech fans and their 4090 tough gaming has dual ball bearing fans that

209
00:14:32,279 --> 00:14:39,480
last up to twice as long as conventional designs it also has capacitors rated for

210
00:14:36,420 --> 00:14:42,300
20 000 hours at 105 degrees Celsius for

211
00:14:39,480 --> 00:14:46,139
added GPU power rail durability and a metal exoskeleton that adds structural

212
00:14:44,040 --> 00:14:50,100
rigidity and vents to increase heat dissipation click the link below and

213
00:14:47,880 --> 00:14:55,139
learn more about ASUS's line of RTX 4000 series gpus today if you guys enjoyed

214
00:14:52,560 --> 00:15:00,120
this video maybe check out our 80 person Studio Tour there are a lot of people

215
00:14:58,199 --> 00:15:05,720
working here now actually there's like four more since then

216
00:15:02,459 --> 00:15:05,720
kind of out of control