WEBVTT 00:00:00.160 --> 00:00:08.000 what do you think can you tell if rtx is on or off going through old posts from 00:00:05.680 --> 00:00:12.400 the launch of the rtx 2000 series is kind of like looking through a war zone 00:00:10.160 --> 00:00:16.480 many gamers and developers were excited about the new visual improvements and 00:00:14.240 --> 00:00:20.720 the ease of development that real-time ray tracing would bring to the table 00:00:18.400 --> 00:00:25.119 others on the other hand were outraged at the idea of paying a premium for a 00:00:22.800 --> 00:00:29.359 feature that they weren't asking for especially if it seemingly came at the 00:00:27.119 --> 00:00:31.279 cost of performance in games that use rasterization 00:00:30.400 --> 00:00:36.559 which at that time was all of them but fast 00:00:34.399 --> 00:00:41.040 forward three years and there are about 50 commercially available games in 00:00:38.399 --> 00:00:44.399 NVIDIA's rtx support list with more to come 00:00:41.840 --> 00:00:47.360 leaving only one question then can normal people even tell the 00:00:45.920 --> 00:00:52.399 difference anyway just like can they tell the difference 00:00:49.680 --> 00:00:55.680 our sponsor glass wire keep track of the weird stuff that's connected to your pc 00:00:54.239 --> 00:01:00.559 even when you aren't using it if a strange device joins your Wi-Fi you'll be 00:00:57.440 --> 00:01:03.760 notified instantly get 25 off using code 00:01:00.559 --> 00:01:03.760 Linus at the link down 00:01:10.840 --> 00:01:16.720 below all right uh 00:01:14.560 --> 00:01:20.320 interesting this doesn't look right trace to me but 00:01:18.080 --> 00:01:22.880 i'm like oh actually that maybe it's gonna be harder than i 00:01:21.280 --> 00:01:27.520 thought oh oh i'm dead 00:01:25.680 --> 00:01:31.520 that's not the test our test setup is pretty straightforward 00:01:29.520 --> 00:01:36.479 jesse from our writing team set up two identical gaming pcs ahead of time so 00:01:33.680 --> 00:01:40.880 that even i won't know which one has rtx on and which one has rtx off our 00:01:38.960 --> 00:01:44.880 subjects get up to five minutes to experience each game and will be 00:01:42.880 --> 00:01:48.799 encouraged to just game as a normal gamer would so they can 00:01:47.280 --> 00:01:53.200 switch back and forth between the two systems instantly using this kvm from 00:01:51.600 --> 00:01:57.119 level one text by the way those guys are awesome and they can do whatever they 00:01:55.280 --> 00:02:01.759 want as long as they don't enable performance metrics and don't manually 00:01:59.920 --> 00:02:04.640 check the settings in the menu because that's cheating can i watch the 00:02:03.600 --> 00:02:10.239 benchmark because there's a scene there at the start oh you can't because one is rtx 00:02:08.319 --> 00:02:13.760 yeah right one's gonna perform way better oh yeah what 00:02:12.400 --> 00:02:17.760 did you learn who's he gonna run a benchmark yeah you can't do that 00:02:16.160 --> 00:02:23.440 i didn't want to look at the FPS it's just once their time is up they simply select 00:02:20.879 --> 00:02:28.570 which system they thought had rtx on and describe what they liked better about 00:02:25.200 --> 00:02:29.760 the rtx experience i've never actually 00:02:29.760 --> 00:02:37.920 seen rtx minecraft before 00:02:33.760 --> 00:02:39.920 where was the lava because i know 00:02:37.920 --> 00:02:46.480 it's supposed to be like extra pretty right yes it's actually like not that hard 00:02:43.280 --> 00:02:46.480 usually if you know what to look for 00:02:46.599 --> 00:02:52.879 um oh sure 00:02:50.800 --> 00:02:56.319 i know i'm so quiet and relaxed i like it Linus always scares me because he 00:02:54.480 --> 00:03:02.480 wants more energy man that bird's got a really long tail 00:02:59.120 --> 00:03:04.080 man that bird's got a really long tail 00:03:02.480 --> 00:03:09.200 while ray tracing is getting a ton of attention from gamers now it's not a new 00:03:06.720 --> 00:03:13.120 idea by any means the basic concept actually predates rasterization and in 00:03:11.280 --> 00:03:18.239 the 3d animation industry it's been in use commercially for well over a decade 00:03:15.599 --> 00:03:21.519 but the thing is in order to determine what a scene should look like using 00:03:20.640 --> 00:03:26.879 real traced light rays you need to emit a 00:03:24.400 --> 00:03:30.879 massive amount of them from a point or ambient light source which will then 00:03:28.879 --> 00:03:35.280 bounce and diffuse around the scene enough times that they eventually become 00:03:33.360 --> 00:03:39.120 sort of negligible bounces and are no longer needed 00:03:36.480 --> 00:03:44.000 now that sounds fine and obvious until you realize that you might need hundreds 00:03:41.120 --> 00:03:48.000 of thousands or millions of rays to even get a somewhat decent picture out of 00:03:46.000 --> 00:03:53.599 them many people don't realize that the simple fact that an image is ray traced 00:03:50.959 --> 00:03:57.680 does not mean that it will look good look at this quick blender render see 00:03:55.519 --> 00:04:01.040 how noisy it looks as though a bunch of the detail is missing 00:03:59.200 --> 00:04:05.760 when non-real-time ray tracing is calculated the work is divided up into 00:04:03.040 --> 00:04:09.599 overlapping chunks called passes and they are added on top of one another 00:04:07.519 --> 00:04:14.879 until a cleanly lit image is produced this can take minutes or even hours for 00:04:12.080 --> 00:04:19.120 a single 4k frame though and that's on modern hardware so you can imagine why 00:04:17.280 --> 00:04:21.759 some new ideas were needed to make this technology viable for games where we 00:04:20.959 --> 00:04:26.880 need 60 frames per second or more while there 00:04:24.880 --> 00:04:30.960 is a clear competitive disadvantage to using ray tracing in a fast-paced title 00:04:29.199 --> 00:04:35.759 like fortnite for example where every frame matters it's still impressive that 00:04:33.840 --> 00:04:41.120 NVIDIA's rt cores work as well as they do in short rt cores are designed to 00:04:38.800 --> 00:04:45.759 perform two tasks that are specific to path tracing which is kind of a shortcut 00:04:43.759 --> 00:04:50.320 version of ray tracing these are called bounding box tests and ray triangle 00:04:48.000 --> 00:04:54.000 intersection tests and they save thousands of instructions per ray that 00:04:52.639 --> 00:04:59.040 would otherwise have to be done by the GPU stream processors and in the 3000 00:04:56.639 --> 00:05:02.400 series their second gen rg cores are more powerful than the first generation 00:05:00.639 --> 00:05:05.759 ones meaning that we should see them continue to improve over time yeah it 00:05:04.320 --> 00:05:08.720 seems like the shadows are just struggling more in this one i don't know 00:05:07.360 --> 00:05:14.160 which one it is because last time i thought the one that looked better in wolfenstein i thought the one that 00:05:11.360 --> 00:05:18.560 looked better was the the rasterized one or the uh the reflection mapped one 00:05:17.280 --> 00:05:21.919 so maybe it's the same thing here where the shadows that look better to me are 00:05:20.479 --> 00:05:28.160 actually the non-raytrace ones the ones that we fake so yeah this is performing better but 00:05:25.600 --> 00:05:32.960 you know it doesn't really it's not that big a deal i guess the problem is that 00:05:30.479 --> 00:05:36.880 it looks better too so 00:05:34.160 --> 00:05:40.800 i mean minecraft beta is beta right so yeah 00:05:39.199 --> 00:05:45.600 it's not like they can't add things later but man like 00:05:43.440 --> 00:05:50.240 that water though just looks so much better so gorgeous it looks like water 00:05:48.479 --> 00:05:55.120 whereas the water in minecraft rtx kind of looks like ocean water all the time 00:05:52.320 --> 00:06:00.560 okay wait a minute if ray tracing is so accurate and therefore better 00:05:57.759 --> 00:06:05.680 what was that how could any test subject fail to tell the difference or worse 00:06:03.120 --> 00:06:09.919 prefer the traditionally rendered method think of raytracing as a final boss 00:06:08.080 --> 00:06:14.000 that's impossible to beat because that's what it was for years 00:06:11.680 --> 00:06:18.720 what's a gamer to do right cheat now obviously we're talking game 00:06:16.479 --> 00:06:24.000 developers so we're i don't mean getting a subscription to game pro magazine here 00:06:20.960 --> 00:06:26.479 but rather i mean building techniques of 00:06:24.000 --> 00:06:31.160 clever tricks that allow you to fake realistic rendering these all together 00:06:29.120 --> 00:06:35.199 have come to be known as rasterization 00:06:33.039 --> 00:06:40.319 raster rendering means that a lot of the lighting shadows reflections and other 00:06:37.600 --> 00:06:44.479 visual elements are baked into the level geometry and handled with packages of 00:06:42.240 --> 00:06:48.240 processing and effects on a per pixel level these are called shaders 00:06:46.720 --> 00:06:53.280 over the years these techniques have become more complex and refined allowing 00:06:50.479 --> 00:06:57.600 things like dynamic shadows in real time by using clever projection mapping as 00:06:55.759 --> 00:07:03.120 well as tricks like more advanced dynamic reflection mapping but the core 00:07:00.639 --> 00:07:06.560 idea of tricking the player into thinking that the lighting is being 00:07:04.479 --> 00:07:10.880 handled in real time has remained the same throughout all of these innovations 00:07:09.039 --> 00:07:14.319 god this is like really hard to tell the difference 00:07:12.160 --> 00:07:21.280 and i i don't really feel like there's like a huge performance drop like what 00:07:17.120 --> 00:07:23.280 GPU are we using 30 80 or 30 90 30 80s 00:07:21.280 --> 00:07:28.000 so i'm like right now creating a hard shadow like harsh like a pretty direct 00:07:25.759 --> 00:07:30.960 reflection of the sun on here on the water 00:07:28.960 --> 00:07:34.319 and let's switch back to system number one 00:07:32.160 --> 00:07:36.960 they look pretty similar i don't really spot a different right 00:07:36.080 --> 00:07:41.759 now as you can see these cheats or tricks 00:07:39.199 --> 00:07:46.960 that game developers use have become so good that they're kind of an obstacle to 00:07:44.639 --> 00:07:50.160 real-time ray tracing catching on even though that final boss is actually 00:07:49.199 --> 00:07:54.000 beatable reflection maps are an excellent example 00:07:52.319 --> 00:07:59.280 of this fake it till you make it approach they're essentially a 360 00:07:57.199 --> 00:08:03.199 degree view of the current scene that's been remapped and projected onto an 00:08:01.520 --> 00:08:08.160 object or surface then with the right amount of distortion 00:08:05.759 --> 00:08:12.639 and selective color removal as well as dynamically swapping out or changing the 00:08:10.319 --> 00:08:16.879 map as you move around you can get high performance visually appealing and quite 00:08:15.039 --> 00:08:20.479 convincing reflections some really good examples of this are 00:08:18.319 --> 00:08:25.039 games like dirt 3 and need for speed hot pursuit it's not perfect of course and 00:08:22.720 --> 00:08:28.400 in games like cs go it's actually very obvious that your scope reflections for 00:08:27.039 --> 00:08:32.719 example don't really match the environment but in the 00:08:30.400 --> 00:08:35.120 heat of a clutch round it's not something that you're likely to notice 00:08:34.320 --> 00:08:39.440 and certainly is a lot better than giving up 00:08:37.599 --> 00:08:43.519 so many FPS leading us finally to the results we've 00:08:41.599 --> 00:08:46.800 been teasing okay so 00:08:44.720 --> 00:08:52.399 for shadow of the tomb raider i believe number two has rtx on what the 00:08:50.240 --> 00:08:56.640 what what do you mean 00:08:54.399 --> 00:09:00.399 i had a feeling i'm like like it's like i'm watching 00:08:57.760 --> 00:09:03.680 like a scientist at work like analyzing okay what is it 00:09:02.160 --> 00:09:09.120 okay so the thing that tipped me off the most is that the shadows in the center of the 00:09:06.320 --> 00:09:13.839 screen here for this tree are shimmering as they're moving 00:09:10.560 --> 00:09:16.080 on the machine with no rtx that 00:09:13.839 --> 00:09:20.320 indicates to me that that's a shadow map whereas the rtx machine 00:09:18.560 --> 00:09:24.240 it's kind of sort of shimmery but it's not 00:09:22.000 --> 00:09:29.360 quite as bad and also the shadows are a little bit softer and they tend to um 00:09:27.519 --> 00:09:34.640 they start sharp and then they go soft whereas the uh shadow map is 00:09:31.760 --> 00:09:39.600 consistently soft you are 100 correct okay as expected then our most technical 00:09:37.200 --> 00:09:44.480 staff had no trouble identifying both the benefits and drawbacks of real-time 00:09:42.160 --> 00:09:49.680 ray tracing but mathematically speaking most people aren't Anthony so 00:09:47.279 --> 00:09:55.120 what about them okay this might be a pretty good spot if i'm facing the sun 00:09:53.200 --> 00:10:00.560 which is like right there i can see reflection on the water here 00:09:58.480 --> 00:10:04.560 and also on the rock here and let's go back to system number two 00:10:03.600 --> 00:10:10.480 wow they look pretty similar 00:10:06.959 --> 00:10:13.600 oh my god yeah i think for number two 00:10:10.480 --> 00:10:14.640 it looks better 00:10:13.600 --> 00:10:20.480 it's like glowier the lava specifically 00:10:19.120 --> 00:10:25.519 there's this effect around it 00:10:22.560 --> 00:10:29.920 yeah whereas like for number four i feel like everything's a bit flatter 00:10:27.760 --> 00:10:34.880 there isn't as much like detail to the the lights and stuff 00:10:32.880 --> 00:10:38.560 like that it's just like a general like 00:10:35.920 --> 00:10:43.200 this block has lava so then the other one should have it around too 00:10:40.880 --> 00:10:46.560 right i don't know i could be totally wrong 00:10:44.160 --> 00:10:50.800 i think this one number two is retracing on 00:10:48.720 --> 00:10:54.000 and what makes you think that um the sharpness of the reflections in 00:10:52.959 --> 00:10:57.839 the water looks a little more reflective than in 00:10:56.320 --> 00:11:03.680 the other one and also the shadows because they're softer 00:11:02.320 --> 00:11:07.680 i feel like that's how leaves would actually look in real life it would be a softer 00:11:06.320 --> 00:11:12.079 shadow because they're all together you wouldn't get these crisp jagged edges 00:11:10.720 --> 00:11:15.920 but now that i say that like i'm wondering if the light's shining so hard 00:11:13.920 --> 00:11:20.480 through it that it's creating those jagged edges with rtx on 00:11:18.000 --> 00:11:26.399 i don't really spot a huge difference but my gut telling me system number one 00:11:23.839 --> 00:11:31.440 it's the one with retracing so this one is the one that we're facing 00:11:30.000 --> 00:11:35.120 although we saw varying levels of success with the number of games each 00:11:33.200 --> 00:11:39.040 person got right and which game specifically they got right 00:11:37.200 --> 00:11:43.200 the overall trend from our test is pretty simple the people with any 00:11:41.200 --> 00:11:48.160 knowledge of 3d rendering and especially with first-hand experience gaming on rtx 00:11:45.440 --> 00:11:52.000 at home were way more likely to nail it as for everyone else 00:11:49.680 --> 00:11:56.000 telling the difference actually isn't easy the truth of the matter is that 00:11:54.240 --> 00:12:01.360 with current titles that support ray tracing it's pretty difficult to tell 00:11:58.880 --> 00:12:05.920 whether it's on or off without pixel peeping more on that in a moment 00:12:03.680 --> 00:12:10.880 even when we did prompt people to look closely or it was extremely challenging 00:12:08.720 --> 00:12:14.320 and the results were quite similar and for fun we threw an extra scenario at 00:12:12.880 --> 00:12:19.360 some of our participants where we had the exact same game running without rtx 00:12:17.040 --> 00:12:25.360 in both cases to see if they could find the difference are they both off oh 00:12:23.600 --> 00:12:29.040 oh yeah okay i was like they yeah i could not tell 00:12:27.440 --> 00:12:33.440 the difference really i guess i was trying to find a difference but uh there 00:12:30.720 --> 00:12:38.800 was none okay they couldn't but they also weren't a hundred percent sure 00:12:36.639 --> 00:12:42.959 so as it turns out if a game already looks really good with rasterization our 00:12:41.760 --> 00:12:46.320 brain will in some cases see things that are not 00:12:45.040 --> 00:12:51.040 necessarily there and this happened to some of our 00:12:48.320 --> 00:12:54.959 participants in wolfenstein for example they would notice parts of the lighting 00:12:52.800 --> 00:12:58.720 that looked you know better in this one when the only difference in that game 00:12:56.560 --> 00:13:02.880 was actually in the reflections bottom line then 00:12:59.680 --> 00:13:04.399 is NVIDIA selling snake oil with rtx 00:13:02.880 --> 00:13:08.639 absolutely not what kind of question is that i mean in games like minecraft for 00:13:06.560 --> 00:13:12.720 Windows 10 and the upcoming metro exodus enhanced edition it is clear how much of 00:13:11.040 --> 00:13:16.720 a difference it can make to the the atmosphere of a game 00:13:14.720 --> 00:13:20.959 and in the long term ray tracing is a huge boon to game developers as well 00:13:19.040 --> 00:13:24.959 because it behaves more like how real light does it can save a ton of time 00:13:23.600 --> 00:13:28.880 that would have otherwise been spent setting thousands of uber hacky 00:13:27.200 --> 00:13:34.079 parameters to get a scene to look convincing but 00:13:30.480 --> 00:13:35.040 the tech is still in a relatively early 00:13:34.079 --> 00:13:40.000 state but it will be the future i mean just 00:13:37.440 --> 00:13:43.839 think about it add in your light sources set your diffusion refraction and 00:13:42.000 --> 00:13:47.920 reflection parameters for every surface and bam you're done 00:13:46.320 --> 00:13:52.720 kind of it's a bit more complicated than that and if you want a more detailed explanation we do have an episode over 00:13:50.800 --> 00:13:56.720 on techwiki that goes into more detail but with all of that said it's clear 00:13:54.639 --> 00:14:01.600 from our experiment today that in the right hands rasterization is still a 00:13:59.279 --> 00:14:06.880 potent tool and this industry transition is going to take many years to complete 00:14:04.480 --> 00:14:12.240 so maybe don't feel too bad if you still haven't managed to get your hands on a 00:14:08.560 --> 00:14:13.279 newfangled rtx GPU 00:14:12.240 --> 00:14:17.600 you know what you should get your hands on though your own website making a 00:14:15.600 --> 00:14:21.040 website doesn't have to be hard just use squarespace and you'll be up and running 00:14:19.040 --> 00:14:24.160 in a matter of hours maybe even faster than that they have award-winning 00:14:22.639 --> 00:14:29.440 templates that'll help make your website stand out instead of looking like it's from the 90s and if you're interested in 00:14:27.920 --> 00:14:32.720 how your website's doing they have built-in tools so you can find out what 00:14:31.040 --> 00:14:36.880 you're doing right and what you're doing wrong both our Linus media group and ltx 00:14:35.120 --> 00:14:40.800 expo websites were built quickly using squarespace and i mean quickly they're 00:14:38.880 --> 00:14:45.120 super easy to maintain and if you get stuck making your website they have 24 7 00:14:42.959 --> 00:14:49.680 support that is ready to help you out so head to squarespace.com LTT and get 10 00:14:48.800 --> 00:14:57.600 off if you guys enjoyed this video maybe check out our 5 000 graphics card video 00:14:54.639 --> 00:15:02.959 on the quadro rtx 8000 i mean what could make a GPU worth 00:14:59.839 --> 00:15:05.199 three times more than this 00:15:02.959 --> 00:15:08.199 five times this is a 3080 right well at msrp