WEBVTT 00:00:00.080 --> 00:00:07.600 those who are in the know know that when it comes to measuring the 00:00:03.840 --> 00:00:10.719 performance of a graphics card FPS is 00:00:07.600 --> 00:00:13.280 just part of the story i mean you could 00:00:10.719 --> 00:00:17.520 have all the FPS in the world but if you had an extremely high latency from your 00:00:15.440 --> 00:00:21.760 mouse click to the graphics card processing it outputting it and actually 00:00:19.520 --> 00:00:24.400 seeing the photons hit your eyeballs that wouldn't make for a very good 00:00:23.039 --> 00:00:28.640 gaming experience furthermore efficiency matters if you 00:00:26.800 --> 00:00:33.200 had two graphics cards that performed the same but one of them sucked back 500 00:00:31.439 --> 00:00:36.399 watts and outputs so much heat in your room that you're like 00:00:34.719 --> 00:00:41.280 gaming like this well surely you'd rather have the more 00:00:38.000 --> 00:00:43.040 efficient one now we've tried a variety 00:00:41.280 --> 00:00:47.680 of different methods of measuring these things in the past everything from clamp 00:00:45.040 --> 00:00:52.480 meters for power to makey makeys with high-speed cameras for input latency and 00:00:50.239 --> 00:00:57.520 sometimes it actually worked out okay but more often than not it required more 00:00:54.640 --> 00:01:04.000 setup than we can put into every single GPU on every single test bench 00:01:00.320 --> 00:01:07.280 fortunately that all changes today meet 00:01:04.000 --> 00:01:11.200 NVIDIA's homegrown tools they're calling 00:01:07.280 --> 00:01:13.680 these the ldat and the p cat 00:01:11.200 --> 00:01:19.600 and we're gonna see a if they work and b we're gonna discuss 00:01:17.040 --> 00:01:22.799 whether or not you can trust them glasswire is the tool that shows you 00:01:21.360 --> 00:01:27.040 which apps are slowing down your connection in real time it's used by 00:01:25.119 --> 00:01:31.119 security pros to monitor for malware block bandwidth wasters and detect 00:01:28.640 --> 00:01:35.640 suspicious activity get 25 off using offer code Linus at the link below 00:01:42.000 --> 00:01:48.159 let's start with ldap the latency and 00:01:45.200 --> 00:01:55.759 display analysis tool it's a rather unassuming looking 00:01:51.280 --> 00:01:58.240 logitech g203 prodigy but wait you need 00:01:55.759 --> 00:02:05.280 to look closely here there's actually a second wire coming out of this thing and 00:02:00.719 --> 00:02:09.000 if we follow it down to the end we find 00:02:05.280 --> 00:02:09.000 a photo sensor 00:02:11.200 --> 00:02:16.879 it's embedded in an unassuming looking little 3d printed housing and that wire 00:02:15.280 --> 00:02:21.120 from the mouse plugs right into the front of it a little something like that 00:02:19.440 --> 00:02:25.040 so then how is this supposed to work 00:02:23.280 --> 00:02:30.560 like so when you place this puppy on the screen 00:02:27.599 --> 00:02:34.000 it detects changes in light output so that means we're actually using a 00:02:32.160 --> 00:02:38.800 similar method to something like the one made by leo bodnar but fundamentally 00:02:36.720 --> 00:02:44.720 different in two ways one instead of taking an HDMI input to 00:02:42.480 --> 00:02:48.959 measure display latency we are actually taking a mouse input to measure end to 00:02:48.080 --> 00:02:53.920 end mouse to photon latency 00:02:51.280 --> 00:02:58.879 two it's software controlled from back to front so through that we can set up 00:02:56.400 --> 00:03:02.720 an auto fire mode that is completely unattended as long as something on 00:03:00.879 --> 00:03:06.560 screen changes every time the mouse clicks we can repeat our tests 00:03:04.560 --> 00:03:11.440 indefinitely to get a reading of potentially hundreds or even thousands 00:03:09.200 --> 00:03:16.959 of data points within a single minute of testing giving us an extremely accurate 00:03:14.000 --> 00:03:21.920 result it even lets us see raw sensor data so we can determine where the best 00:03:19.360 --> 00:03:25.599 place on the screen is to place the sensor you guys might not know this but 00:03:23.840 --> 00:03:30.720 screens actually draw like this not just 00:03:28.000 --> 00:03:35.200 like randomly not all at once yeah dead center is fine um ideally what it would 00:03:32.959 --> 00:03:39.200 be is like say you're measuring a game so if it's a first person shooter 00:03:37.519 --> 00:03:42.640 then you would be putting it over the muzzle where the muzzle flash would be 00:03:40.799 --> 00:03:46.720 press the button so why don't we go with my my personal favorite we'll go with 00:03:44.879 --> 00:03:51.120 that that cs go well there's you're missing a step oh i see there was 00:03:48.720 --> 00:03:54.640 another cable in that box yeah and it's still in there oh yeah wait what's this 00:03:52.720 --> 00:03:57.680 for that plugs into the light sensor well let's fire up the program first 00:03:56.319 --> 00:04:00.959 that would be on the desktop it's just in a folder called ldap ldap oh yeah i 00:03:59.760 --> 00:04:06.720 can measure audio as well it's got a little 3.5 millimeter audio jack in there so it can measure audio latency 00:04:05.200 --> 00:04:11.760 in theory you could measure any audio latency with that like not even a computer like 00:04:09.519 --> 00:04:15.360 a tv or something so if you keep it in free running mode then if as you move 00:04:13.760 --> 00:04:20.239 that around uh there you go see there was a little 00:04:18.000 --> 00:04:23.040 luminous change whoa look at that 00:04:21.440 --> 00:04:26.240 so if you put a brighter window over that or some kind of 00:04:26.639 --> 00:04:33.360 yeah like task manager whoa cool yeah so you can actually set the 00:04:31.520 --> 00:04:38.880 activation point for what luminance it should like pick up on uh-huh and uh 00:04:36.720 --> 00:04:42.560 yeah it'll just do it it's super fast sensor 00:04:40.560 --> 00:04:47.040 that's cool you can also add the uh mouse button as 00:04:44.880 --> 00:04:51.040 well with that second option so that it it actually will detect when 00:04:49.199 --> 00:04:54.960 you click the mouse this mode is mostly useful for setting things up the first 00:04:52.720 --> 00:04:58.479 time and that histogram is what you get after you get the latency calculation 00:04:57.040 --> 00:05:03.520 right now if you wanted to get something that's actually a latency calculation 00:05:00.639 --> 00:05:09.720 uh put the window for uh ldap in front of the sensor now click on the window 00:05:05.759 --> 00:05:09.720 like somewhere other than a control 00:05:10.240 --> 00:05:14.880 oh cool yeah so you can just do that it'll pick 00:05:14.000 --> 00:05:19.440 it up how how long it took for it to change so 00:05:17.199 --> 00:05:22.960 what it's doing is it's actually sending the pulse to make the click down the 00:05:21.360 --> 00:05:26.639 cable that's connected to the mouse it's not actually like using Windows or 00:05:25.280 --> 00:05:31.840 anything else to make the click it's it's all in in the hardware so you didn't have to have the software running 00:05:29.759 --> 00:05:35.440 on this computer not only does this automate the process and make it so that 00:05:33.680 --> 00:05:40.720 we can run these tests without a secondary pc or like camera capture 00:05:38.160 --> 00:05:45.520 setup it's going to save so much time it even exports in csv form which we can 00:05:43.600 --> 00:05:48.800 then import into excel and turn into nice little charts for you guys 00:05:47.600 --> 00:05:53.600 but it doesn't tell us 00:05:50.639 --> 00:05:59.840 what is contributing to the overall input lag it just tells us the total so 00:05:57.199 --> 00:06:04.880 what we're measuring here is everything debounce usb polling rate operating 00:06:03.039 --> 00:06:09.360 system processing delay driver processing delay game engine processing 00:06:06.720 --> 00:06:14.800 delay render to display delay and then finally of course the pixel response 00:06:11.440 --> 00:06:17.360 time of our display which isn't to say 00:06:14.800 --> 00:06:20.400 though that it's useless because if we've got this data it means that we can 00:06:19.680 --> 00:06:27.039 change other variables one at a time and just see if the 00:06:23.600 --> 00:06:29.440 overall delay gets better but it's got 00:06:27.039 --> 00:06:34.560 NVIDIA right on the box can we trust it 00:06:31.840 --> 00:06:37.360 yeah i mean it's just a mouse with wires coming out of the button that goes into 00:06:36.160 --> 00:06:40.960 the sensor you don't even have to plug it into the same computer it has nothing 00:06:39.280 --> 00:06:44.400 to do with NVIDIA so yeah i don't see any reason why not 00:06:43.199 --> 00:06:50.960 cool all right how about this other tool this 00:06:47.680 --> 00:06:52.479 is the pcat or power capture analysis 00:06:50.960 --> 00:06:58.160 tool you can tell these are named by the engineers right this is probably the 00:06:54.880 --> 00:07:01.199 most interesting thing i've seen in a 00:06:58.160 --> 00:07:02.720 long time when it comes to measuring 00:07:01.199 --> 00:07:07.599 graphics parts tools to accurately measure power 00:07:05.039 --> 00:07:13.360 consumption are not easily accessible being either very expensive or very rare 00:07:11.440 --> 00:07:19.520 or even worse both methods we've used in the past such as 00:07:16.560 --> 00:07:24.240 kilowatts don't accurately measure power draw because power supply inefficiencies 00:07:21.759 --> 00:07:28.240 can creep in to say nothing of the rest of the system's power consumption during 00:07:26.160 --> 00:07:32.400 the test even worse are the software derived methods you might use at home 00:07:29.919 --> 00:07:36.000 because a each vendor has a different idea of how to measure total board power 00:07:34.720 --> 00:07:40.000 and b you're counting on them to report it correctly 00:07:41.280 --> 00:07:46.639 never go up against a sicilian when death is on the line 00:07:46.960 --> 00:07:56.720 which brings us then to these two pcbs one with a pci express 00:07:53.199 --> 00:08:00.240 connector and card edge and another with 00:07:56.720 --> 00:08:01.360 a whole whack ton of pci express power 00:08:00.240 --> 00:08:06.319 connectors as you guys might have guessed 00:08:03.440 --> 00:08:09.919 these allow us to measure the power flowing through each of these connectors 00:08:08.879 --> 00:08:15.440 that is both from the power supply cables that 00:08:12.720 --> 00:08:21.039 come into the graphics card and through the motherboard itself with up to three 00:08:17.840 --> 00:08:23.599 eight pin pci express 12v cables 00:08:21.039 --> 00:08:28.879 now on this puppy right here we've got a removable OLED display that takes both 00:08:26.080 --> 00:08:32.880 the feed from the pci express board and the power interposers and spits out 00:08:30.800 --> 00:08:38.080 statistics and the current power draw NVIDIA claims it's within 0.25 of their 00:08:35.919 --> 00:08:43.599 oscilloscopes as far as accuracy goes so with this we can accurately measure what 00:08:40.640 --> 00:08:49.440 the real world power draw is for any pci express card actually not just a 00:08:47.279 --> 00:08:53.120 graphics card let's go ahead and try it out here this is one of those things 00:08:50.959 --> 00:08:57.360 that as a manufacturer you have to have the absolute utmost confidence in your 00:08:55.440 --> 00:09:00.560 products in order to produce because it's the kind of thing that 00:08:59.120 --> 00:09:04.080 obviously both AMD and NVIDIA have had internally 00:09:02.800 --> 00:09:09.120 for years but if you go ahead and you make all these 00:09:06.800 --> 00:09:14.000 tools available to you know every yahoo you know hardware reviewer out there 00:09:11.360 --> 00:09:17.279 then not just today but for years forever to come they're going to have 00:09:15.680 --> 00:09:20.880 the means to hold you accountable for your crappy products so you basically 00:09:18.800 --> 00:09:24.240 have to think you know not only do i have the best product 00:09:22.560 --> 00:09:29.040 today but i'm gonna have the best product forever and i want everyone to 00:09:26.320 --> 00:09:32.720 know about it to produce tools like this and send them out because i've asked for 00:09:30.560 --> 00:09:36.000 years like you guys obviously have all these measurement tools why don't you 00:09:34.399 --> 00:09:40.080 just provide them to us because you have them anyway and the answer is always 00:09:37.760 --> 00:09:43.440 like uh let me get back to you 00:09:42.320 --> 00:09:48.959 and yet here we are this is freaking awesome 00:09:47.040 --> 00:09:52.080 how replaceable is this Anthony is this just something NVIDIA's going to sell on 00:09:50.399 --> 00:09:55.760 their webstore or like is it reviewers only or for now it's a 00:09:54.240 --> 00:10:00.720 reviewers-only kind of thing it's just something that they've kind of built upon out of their lab that's why you see 00:09:58.880 --> 00:10:04.560 the 3d printed components but they haven't ruled out selling it to 00:10:02.640 --> 00:10:08.320 consumers if there's enough demand for it so uh 00:10:06.320 --> 00:10:11.279 maybe let them know down in the comments if you want to see that just just make 00:10:09.920 --> 00:10:17.200 sure that they're on the input side versus the other output side and there's 00:10:13.600 --> 00:10:18.640 a wire on each end all right so 00:10:17.200 --> 00:10:22.560 let's just put that puppy i don't know like 00:10:21.200 --> 00:10:29.600 brilliant so they include up to a triple slot 00:10:26.000 --> 00:10:31.600 of little dude adama jigs that you can 00:10:29.600 --> 00:10:35.600 screw in here to your case so you can actually mount it securely apparently 00:10:33.360 --> 00:10:38.160 it's a standard item for mcmaster car you can actually just order them 00:10:36.800 --> 00:10:42.640 yourself if you need a different threading for your screws oh okay 00:10:40.720 --> 00:10:44.959 uh well which i think may actually be necessary in our case all right we're 00:10:44.160 --> 00:10:50.720 just huh we're going straight yolo here boys 00:10:47.600 --> 00:10:53.519 oh yeah no problem piece of cake 00:10:50.720 --> 00:10:56.880 oh the display is upside down this is why we didn't give Linus the 00:10:54.959 --> 00:11:01.920 tools okay okay here we go total power 14 watts 00:11:00.240 --> 00:11:06.560 right now dang oh i should plug in an HDMI cable 00:11:04.959 --> 00:11:11.120 yeah power usage probably go up a little bit with the HDMI cable in yeah the good 00:11:09.120 --> 00:11:15.680 news is the HDMI cable should anchor it to the back of the case here so oh 00:11:13.120 --> 00:11:18.959 that's not a good solution what do you what do you work at Linus 00:11:17.040 --> 00:11:23.360 tech tips or something what's wrong with you listen we deal in jank around here 00:11:21.200 --> 00:11:27.920 right of course you're not stuck just looking at this little tiny display down 00:11:25.360 --> 00:11:34.079 here as freaking cool as that is there's an accompanying pcat utility v 1.1 that 00:11:31.279 --> 00:11:39.680 like the ldap one can log to csv files and additionally provide real-time power 00:11:36.399 --> 00:11:42.480 graphs look at this not just total but 00:11:39.680 --> 00:11:46.079 on a per-source basis so on our two eight pins here you can 00:11:44.880 --> 00:11:51.360 see we're doing somewhere in the neighborhood of four 00:11:48.640 --> 00:11:54.720 watts or so and then where is it yep 12 volt pci express so 00:11:53.600 --> 00:11:59.040 that's coming through the slot we're doing about four or five watts so it's actually equally balanced across all of 00:11:58.320 --> 00:12:03.040 them so then i can just fire up furmark right 00:12:01.680 --> 00:12:06.959 yeah and if you wanted to have all of those different uh metrics on the graph 00:12:05.440 --> 00:12:10.320 there was actually check boxes down at the bottom right there and that'll just 00:12:08.480 --> 00:12:15.760 throw them on there for you even just the like the Windows 00:12:12.399 --> 00:12:17.760 hardware GPU acceleration of the ui 00:12:15.760 --> 00:12:21.279 will give you these little spikes when you do stuff like watch i'm going to 00:12:19.200 --> 00:12:26.600 launch fur mark and it's like oh 40 watts hey we got to draw this window are 00:12:23.279 --> 00:12:29.120 you okay everybody calm down now 00:12:26.600 --> 00:12:33.440 250 watts and i guess we can 00:12:31.200 --> 00:12:35.920 believe it because that's about what we would expect out of a board like this 00:12:35.200 --> 00:12:42.480 and i mean this is just this is just power wires like it would 00:12:39.680 --> 00:12:48.320 have no way whatsoever of knowing if we have a Radeon or a g-force or a 00:12:45.760 --> 00:12:48.320 Intel 00:12:49.120 --> 00:12:54.560 not yet but later you know at some point yeah it's completely agnostic as far as 00:12:52.560 --> 00:12:58.480 it's concerned it can just measure power from whatever source doesn't matter if 00:12:56.240 --> 00:13:00.959 it's a graphics card or a sound card or a network card 00:12:59.680 --> 00:13:05.920 and just figure out what the power draw is we are going to be able to do the coolest like power draw charts for you 00:13:04.959 --> 00:13:11.120 guys and it gets even better because finally both 00:13:09.200 --> 00:13:17.839 of these tools integrate into frame view 1.1 to show latency and frames dropped 00:13:14.800 --> 00:13:20.240 by the display and power statistics from 00:13:17.839 --> 00:13:24.800 pcat now that the overlay has vulcan support plus we've got integration of 00:13:22.560 --> 00:13:29.519 these other tools there's actually very little that we wish that it did 00:13:27.040 --> 00:13:34.320 that it doesn't already directx 9 and 10 overlay next please and thank you 00:13:32.720 --> 00:13:41.519 let's go ahead and take a crack at this campaign so we've jumped to 4k and we're gonna 00:13:38.720 --> 00:13:47.320 see how this puppy performs here how much power it draws 00:13:43.519 --> 00:13:47.320 we're gonna know for sure 00:13:48.800 --> 00:13:56.800 broken ARM thing in your mouth boom broken head 00:13:54.880 --> 00:14:00.079 you fell off a cliff take that 00:13:58.160 --> 00:14:03.440 oh that guy threw a thing at me don't do that anyway that's not the point today 00:14:02.240 --> 00:14:08.000 the point is we are sucking back around 00:14:05.399 --> 00:14:11.360 250 water rooney's with this thing oh wow it's got right it's got the little 00:14:09.600 --> 00:14:14.399 display on it i don't even have to like all tab out i can see it i can see it 00:14:13.199 --> 00:14:19.279 right there i love this this is this is freaking 00:14:17.120 --> 00:14:22.320 cool and i'm sure that if the price is reasonable a lot of enthusiasts are 00:14:21.040 --> 00:14:26.399 going to want to get their hands on these things come on NVIDIA put it on 00:14:24.560 --> 00:14:31.120 the store today's video is brought to you by the drop objective 2 headphone 00:14:28.160 --> 00:14:35.680 amplifier aka the o2 amp this amp was designed and created with the feedback 00:14:33.279 --> 00:14:40.000 from over 500 drop members on what they wanted and serves as an ideal baseline 00:14:37.920 --> 00:14:44.880 reference amplifier powering everything from in-ear monitors to hd800s you can 00:14:42.800 --> 00:14:49.440 adjust the input output and power arrangements in two different gains 00:14:46.800 --> 00:14:53.920 medium and standard and it delivers big clear and accurate sound check it out at 00:14:52.000 --> 00:14:59.480 the link down below okay 00:14:55.040 --> 00:14:59.480 um what are we thinking for thumbnail