WEBVTT 1 00:00:00.120 --> 00:00:04.320 When you walk into a massive, sprawling theme park. The 2 00:00:04.360 --> 00:00:08.359 goal is total immersion, Like do you take in the 3 00:00:08.400 --> 00:00:13.880 towering castles, the incredibly lifelike animatronics, the fireworks perfectly sync 4 00:00:13.960 --> 00:00:17.359 to the orchestral swells, and you just sort of accept it, right, 5 00:00:17.399 --> 00:00:20.519 You don't really question it exactly. The illusion works provided 6 00:00:20.559 --> 00:00:22.039 you don't look down, because. 7 00:00:21.800 --> 00:00:24.280 If you look down, I mean, you'd be confronted with 8 00:00:24.519 --> 00:00:28.359 miles of subterranean utility tunnels. You'd see the high pressure 9 00:00:28.399 --> 00:00:32.759 hydraulic pumps, the massive server farms timing well, every single 10 00:00:32.840 --> 00:00:37.479 sparkle of light. Yeah, the magic relies entirely on this relentless, 11 00:00:37.759 --> 00:00:41.320 hyperstructured mechanical reality running right beneath your feet. 12 00:00:41.600 --> 00:00:44.560 And today we are going down into those punnels. Welcome 13 00:00:44.600 --> 00:00:48.799 to the deep dive. We are exploring the invisible, hypercomplex 14 00:00:48.840 --> 00:00:52.719 software frameworks that run today's multi billion dollar digital worlds. 15 00:00:52.960 --> 00:00:54.759 We are talking about game engines. 16 00:00:54.799 --> 00:00:55.960 It's a huge topic, it. 17 00:00:55.960 --> 00:00:59.119 Really is, and we're basing this analysis on a brilliantly 18 00:00:59.119 --> 00:01:01.920 detailed source. Today excerpts from the third edition of the 19 00:01:01.920 --> 00:01:05.079 textbook Game Engine Architecture by Jason Gregory, which came out 20 00:01:05.079 --> 00:01:05.879 in twenty nineteen. 21 00:01:06.040 --> 00:01:08.400 It is essentially the bible for this stuff totally. 22 00:01:08.719 --> 00:01:11.519 Our mission for you today is to demystify what a 23 00:01:11.560 --> 00:01:15.519 game engine actually is, how it orchestrates literally everything from 24 00:01:15.640 --> 00:01:19.280 you rendering a single beam of refracted light to simulating 25 00:01:19.319 --> 00:01:22.359 the gravity of an entire planet. And we'll look at 26 00:01:22.400 --> 00:01:26.439 why these massive digital infrastructures are ultimately custom tailored to 27 00:01:26.599 --> 00:01:28.239 hack the human visual system. 28 00:01:28.319 --> 00:01:31.719 Yeah, because it's an industry built on art and narrative, right, Yeah, 29 00:01:31.760 --> 00:01:36.319 But beneath the surface esthetics, these engines are ultimately highly 30 00:01:36.359 --> 00:01:38.760 sophisticated mathematical models of reality. 31 00:01:38.959 --> 00:01:39.280 Right. 32 00:01:39.400 --> 00:01:44.920 They are just gargantuan numerical simulations operating under immense computational constraints. 33 00:01:45.120 --> 00:01:48.280 Okay, let's unpack this because the concept of a game 34 00:01:48.280 --> 00:01:51.719 engine as a standalone piece of software, it wasn't always 35 00:01:51.760 --> 00:01:52.079 a thing. 36 00:01:52.239 --> 00:01:52.879 No, not at all. 37 00:01:52.959 --> 00:01:55.840 If you look at the architecture of early console eras 38 00:01:55.920 --> 00:01:58.599 like the Intellivision or the Atari, there was no engine. 39 00:01:59.239 --> 00:02:02.519 The source material emphasizes that these early games were completely 40 00:02:02.560 --> 00:02:03.760 monolithic blocks of. 41 00:02:03.719 --> 00:02:06.000 Coupe written directly in assembly right. 42 00:02:06.000 --> 00:02:09.800 Directly addressing the specific silicon of the hardware. The game 43 00:02:09.879 --> 00:02:12.639 and the hardware were essentially one fused entity. 44 00:02:13.039 --> 00:02:16.840 And the memory constraints were so incredibly tight that programmers 45 00:02:16.840 --> 00:02:20.639 had to manually flip individual memory registers just to draw 46 00:02:20.680 --> 00:02:22.960 a single pixel. Oh wow, Yeah, you couldn't just take 47 00:02:23.000 --> 00:02:25.520 the codebase for a maze game like pac Man and 48 00:02:25.599 --> 00:02:29.120 poured it over to a space shooter. The logic governing 49 00:02:29.120 --> 00:02:32.039 how a ghost moved was hopelessly tangled up with the 50 00:02:32.080 --> 00:02:37.039 actual hardware interrupt telling the cathode raypube to draw the color. 51 00:02:36.840 --> 00:02:39.840 Red, which sounds like a nightmare for reusability. But the 52 00:02:39.919 --> 00:02:43.759 nineteen nineties kind of shattered that monolithic approach, largely spearheaded 53 00:02:43.800 --> 00:02:46.840 by a John Carmack and AID Software with Doom and Quake. 54 00:02:47.039 --> 00:02:49.120 Yes, absolute revolutionary stuff. 55 00:02:49.159 --> 00:02:52.639 What they did was architecturally massive. They decoupled the core 56 00:02:52.719 --> 00:02:56.879 software components, the three D graphics renderer, the audio spatialization, 57 00:02:57.719 --> 00:03:00.719 the physics from the art assets, and the secific rules 58 00:03:00.759 --> 00:03:01.639 of the game itself. 59 00:03:01.879 --> 00:03:04.280 And that separation of concerns didn't just clean up the 60 00:03:04.319 --> 00:03:07.879 code base, it accidentally birthed the entire modern community. 61 00:03:08.159 --> 00:03:08.280 Right. 62 00:03:08.439 --> 00:03:12.080 Players suddenly realized they could swap out the demons for 63 00:03:12.159 --> 00:03:16.240 aliens while keeping all the foundational physics and rendering intact. 64 00:03:15.960 --> 00:03:18.439 Which brings us to a core concept the author calls 65 00:03:18.520 --> 00:03:22.240 the reusability gamut because on one extreme of this spectrum 66 00:03:22.280 --> 00:03:26.520 you have the highly specific, single use code of the Arcade. 67 00:03:26.159 --> 00:03:28.319 Era right the glued together so yeah. 68 00:03:28.360 --> 00:03:30.439 And in the middle you have early engines like the 69 00:03:30.479 --> 00:03:34.520 original Quake framework. Those could be heavily modified, but realistically, 70 00:03:34.599 --> 00:03:36.280 I mean, you're really only going to use it to 71 00:03:36.319 --> 00:03:37.879 make other first person shooters. 72 00:03:37.960 --> 00:03:40.560 You weren't making a puzzle game with the Quake engine exactly. 73 00:03:40.759 --> 00:03:42.599 But on the far end of the spectrum you have 74 00:03:42.719 --> 00:03:47.280 today's highly reusable platforms like Unreal Engine or Unity, which 75 00:03:47.319 --> 00:03:51.479 are engineered to build virtually any interactive experience you can imagine. 76 00:03:51.520 --> 00:03:54.120 And reaching that far end of the gamut relies entirely 77 00:03:54.319 --> 00:03:58.520 on data driven architecture, meaning a true modern engine doesn't 78 00:03:58.520 --> 00:04:02.000 have a concept of say, a wizard or a racecar 79 00:04:02.439 --> 00:04:05.280 hard coded into it C plus plus rendering loop. 80 00:04:05.439 --> 00:04:06.280 Okay, so what does it have? 81 00:04:06.479 --> 00:04:09.080 Well, the engine is just a blind, hyper efficient factory 82 00:04:09.120 --> 00:04:13.800 for drawing polygons and calculating light bounces. The specific wizardness 83 00:04:13.800 --> 00:04:17.199 of a character is defined entirely externally by data files 84 00:04:17.279 --> 00:04:19.560 and scripts that get fed into the engine at runtime. 85 00:04:19.839 --> 00:04:22.040 So to use an analogy. Building a game in the 86 00:04:22.079 --> 00:04:25.519 early eighties was basically like building a house where the furniture, 87 00:04:25.639 --> 00:04:29.040 the wallpaper, and the appliances are permanently glued to the 88 00:04:29.040 --> 00:04:29.639 floor plan. 89 00:04:29.879 --> 00:04:31.079 Huh, yes, exactly. 90 00:04:31.160 --> 00:04:33.279 If you want a new couch, you literally have to 91 00:04:33.319 --> 00:04:36.519 tear down the load bearing walls. But a modern game 92 00:04:36.560 --> 00:04:40.240 engine is like buying a hyper advanced prefab foundation. You 93 00:04:40.279 --> 00:04:43.000 get the plumbing, the electrical grid, and the physical laws 94 00:04:43.000 --> 00:04:45.120 of gravity right out of the box, and you just 95 00:04:45.199 --> 00:04:47.959 swap out the drywall and the decor at will. 96 00:04:48.199 --> 00:04:51.959 That's spot on the engine is just the utility grid. 97 00:04:52.000 --> 00:04:55.240 It doesn't care if it's powering a toaster or a supercomputer. 98 00:04:55.720 --> 00:04:58.560 It just routes the electricity flawlessly. 99 00:04:58.839 --> 00:05:01.680 But I'd assume that if you unity or unreal handles 100 00:05:01.720 --> 00:05:04.480 all that underlying plumbing out of the box, building a 101 00:05:04.519 --> 00:05:08.000 game should theoretically be leaner today, right you would think so? Yeah, yeah. 102 00:05:08.040 --> 00:05:10.800 When you look at modern game development studios require these 103 00:05:10.879 --> 00:05:15.160 multidisciplinary armies of hundreds of engineers and artists. If the 104 00:05:15.160 --> 00:05:17.920 foundation is already built for them, why has the workforce 105 00:05:17.959 --> 00:05:19.360 scale up so dramatically. 106 00:05:19.680 --> 00:05:23.600 What's fascinating here is how modularity actually triggered an arms 107 00:05:23.680 --> 00:05:27.879 race and fidelity because the engine is so incredibly capable. Now, 108 00:05:28.240 --> 00:05:31.240 the sheer volume of data required to make a modern 109 00:05:31.279 --> 00:05:34.800 digital world feel real to you is astronomical. 110 00:05:34.879 --> 00:05:37.399 Oh, I see, it's the content itself that's bloated. 111 00:05:37.600 --> 00:05:42.000 Exactly. You no longer have one programmer meticulously drawing pixels. 112 00:05:42.279 --> 00:05:46.160 You have a massive structural divide in the engineering department alone. 113 00:05:46.240 --> 00:05:49.759 You have runtime engineers writing the high performance C plus 114 00:05:49.800 --> 00:05:52.120 plus that executes while you play. 115 00:05:52.240 --> 00:05:54.000 And then you have tools engineers, right. 116 00:05:54.000 --> 00:05:57.240 Yes, tools engineers whose entire job is just building the 117 00:05:57.279 --> 00:05:59.720 custom plugins that allow the artists to even get their 118 00:05:59.720 --> 00:06:01.720 masses of files into the engine in the first place. 119 00:06:01.839 --> 00:06:02.800 That's wild. 120 00:06:03.079 --> 00:06:08.120 The art pipeline alone requires concept artists, photogrammetry specialists, rigging engineers, 121 00:06:08.199 --> 00:06:12.000 motion capture technicians. I mean, the engine is a bottomless furnace, 122 00:06:12.319 --> 00:06:16.000 and it requires a tremendously complex human organization just to 123 00:06:16.040 --> 00:06:18.079 shovel enough high quality data into it. 124 00:06:18.199 --> 00:06:21.480 So you have this massive team feeding high fidelity assets 125 00:06:21.519 --> 00:06:25.800 into a modular foundation, but a beautifully modeled million polygon 126 00:06:25.920 --> 00:06:29.160 character as well. It's entirely useless if it's static. 127 00:06:28.920 --> 00:06:31.040 Right, It's just a digital statue. 128 00:06:30.519 --> 00:06:33.519 Exactly, the engine has to bring it to life, which 129 00:06:33.519 --> 00:06:37.839 introduces the ultimate constraint time. The textbook uses a highly 130 00:06:37.879 --> 00:06:41.959 specific phrase to define this environment. It calls modern video 131 00:06:42.000 --> 00:06:47.839 games soft real time interactive agent based computer simulations, and. 132 00:06:47.879 --> 00:06:50.959 Every single modifier in that phrase is doing a lot 133 00:06:50.959 --> 00:06:51.759 of heavy lifting. 134 00:06:51.879 --> 00:06:53.240 Break that down for us well. 135 00:06:53.360 --> 00:06:55.800 To understand the simulation aspect, we really have to look 136 00:06:55.800 --> 00:06:59.199 at the math. In applied mathematics, you generally deal with 137 00:06:59.279 --> 00:07:03.680 either analyst models or numerical models. An analytic model can 138 00:07:03.720 --> 00:07:05.759 be solved mathematically in a closed. 139 00:07:05.439 --> 00:07:07.279 Form, meaning you can predict it perfectly. 140 00:07:07.519 --> 00:07:10.120 Right. If I fire a cannon ball in a vacuum, 141 00:07:10.439 --> 00:07:13.199 I can plug time into a standard ballistic equation and 142 00:07:13.240 --> 00:07:16.040 tell you the exact coordinates of that cannonball at any 143 00:07:16.040 --> 00:07:17.160 millisecond in the future. 144 00:07:17.600 --> 00:07:20.360 But a game engine can't rely on closed form equations 145 00:07:20.439 --> 00:07:23.360 because of the interactive part of that definition. You have 146 00:07:23.439 --> 00:07:25.519 no idea when the player is going to suddenly apply 147 00:07:25.600 --> 00:07:28.279 the brakes, fire a weapon, or just you know, crash 148 00:07:28.279 --> 00:07:28.959 a car into a. 149 00:07:28.920 --> 00:07:32.959 Wall precisely, which is why games rely on numerical models. 150 00:07:33.480 --> 00:07:36.439 They utilize a main game loop, which is essentially just 151 00:07:36.639 --> 00:07:40.040 a massive wild loop running continuously in. 152 00:07:40.000 --> 00:07:42.240 The background just constantly looping. 153 00:07:42.120 --> 00:07:45.319 Over and over. During every single iteration of that loop, 154 00:07:45.480 --> 00:07:49.120 the engine calculates a delta time, which is the tiny 155 00:07:49.199 --> 00:07:51.959 fraction of a second since the last loop, and uses 156 00:07:52.079 --> 00:07:55.360 numerical integration to step the entire state of the world 157 00:07:55.480 --> 00:07:59.279 forward by that exact margin. It's reading player inputs and 158 00:07:59.399 --> 00:08:00.959 updating THEVLA on the fly. 159 00:08:01.439 --> 00:08:03.279 So what does this all mean for the listener? If 160 00:08:03.279 --> 00:08:05.959 I'm visualizing the game loop, it's almost like a wildly 161 00:08:06.040 --> 00:08:09.639 complex flip book animation. But the frames aren't pre drawn. 162 00:08:09.800 --> 00:08:11.720 No, they're completely blank. 163 00:08:11.560 --> 00:08:14.639 Right, They're being sketched in real time by this hyperactive 164 00:08:14.680 --> 00:08:18.639 mathematician who has to recalculate the physics, the lighting, the 165 00:08:18.680 --> 00:08:22.120 audio propagation, and the AI behavior of an entire universe 166 00:08:22.160 --> 00:08:23.439 before turning the page. 167 00:08:23.519 --> 00:08:26.639 And that mathematician is working under the tyranny of the deadline, 168 00:08:26.680 --> 00:08:30.439 which is the defining characteristic of any real time system, because. 169 00:08:30.199 --> 00:08:32.240 If the monitor refreshes and the frame isn't. 170 00:08:32.000 --> 00:08:36.320 Ready, the illusion shatters. For the cognitive illusion of motion 171 00:08:36.440 --> 00:08:39.679 to hold, the game loop must finish all its calculations 172 00:08:39.759 --> 00:08:42.000 before the monitor is ready to draw the next frame. 173 00:08:42.679 --> 00:08:45.799 If a game targets sixty frames, per second, the engine 174 00:08:45.840 --> 00:08:50.759 has exactly sixteen point six milliseconds to calculate the entire universe. 175 00:08:50.840 --> 00:08:54.720 Sixteen point six milliseconds. That is barely any time at all. 176 00:08:54.759 --> 00:08:59.360 And to complicate matters, the engine is orchestrating overlapping deadlines asynchronously. 177 00:09:00.039 --> 00:09:02.840 The graphics pipeline needs to update sixty times a second, 178 00:09:03.120 --> 00:09:05.200 but the physics engine might be running at one hundred 179 00:09:05.240 --> 00:09:08.320 and twenty hertz just to ensure fast moving objects don't 180 00:09:08.320 --> 00:09:09.600 clip through walls. 181 00:09:09.360 --> 00:09:11.759 Right otherwise, a fast bullet might pass right through a 182 00:09:11.799 --> 00:09:13.720 thin wall between frames exactly. 183 00:09:14.320 --> 00:09:17.120 Meanwhile, the aipath finding might only need to update twice 184 00:09:17.159 --> 00:09:20.000 a second, so the engine is essentially a master conductor, 185 00:09:20.120 --> 00:09:23.559 keeping all these distinct tempos perfectly synchronized. 186 00:09:23.080 --> 00:09:25.879 Which brings us to the distinction between soft and hard 187 00:09:26.080 --> 00:09:30.039 real time systems. Video games operate in soft real time, 188 00:09:30.480 --> 00:09:34.039 meaning if our hyperactive mathematician misses that sixteen point six 189 00:09:34.080 --> 00:09:37.000 millisecond deadline, the system doesn't completely crash. 190 00:09:37.399 --> 00:09:41.080 No, the game simply drops a frame, the animation stutters, 191 00:09:41.279 --> 00:09:44.720 the player gets slightly annoyed, but you know, everyone lives. 192 00:09:44.879 --> 00:09:48.000 Whereas in a hard real time system, a miss deadline 193 00:09:48.200 --> 00:09:49.399 is literally fatal. 194 00:09:49.639 --> 00:09:54.039 Yes, If the avionic software stabilizing a helicopter or the 195 00:09:54.120 --> 00:09:57.799 automated control rod mechanism and a nuclear reactor takes twenty 196 00:09:57.879 --> 00:10:02.440 milliseconds to execute instead of sixteen, the system fails catastrophically. 197 00:10:02.679 --> 00:10:05.759 So games have the luxury of being soft real time, 198 00:10:06.120 --> 00:10:11.240 Yet developers still bend their entire architectural strategy around never missing. 199 00:10:10.960 --> 00:10:14.759 That deadline because the human visual cortex is just agonizingly 200 00:10:14.799 --> 00:10:18.039 sensitive to the illusion breaking hit the deadline and your 201 00:10:18.039 --> 00:10:22.639 brain perceives discrete numerical calculations as a continuous, breathing reality. 202 00:10:22.759 --> 00:10:26.720 And hitting those deadlines consistently while simulating millions of polygons 203 00:10:26.720 --> 00:10:30.639 and complex physics is computationally impossible. If you try to 204 00:10:30.639 --> 00:10:33.720 render reality perfectly, Engines have to cheat. 205 00:10:33.879 --> 00:10:35.000 Oh, they cheat constantly. 206 00:10:35.120 --> 00:10:37.720 The architecture of the illusion is highly dependent on the 207 00:10:37.720 --> 00:10:40.679 specific genre constraints of the experience you're building. 208 00:10:40.879 --> 00:10:45.039 You are entirely governed by trade offs. The algorithmic shortcuts 209 00:10:45.080 --> 00:10:48.120 you take depend entirely on what the player is actually 210 00:10:48.200 --> 00:10:50.279 expected to do in that specific game. 211 00:10:50.519 --> 00:10:53.720 Take a traditional first person shooter, for example, you're demanding 212 00:10:53.799 --> 00:10:58.279 hyperrealistic rendering, but structurally you are usually moving the camera 213 00:10:58.320 --> 00:11:00.720 down in closed corridors or type site lines. 214 00:11:00.840 --> 00:11:02.519 Right, you can't see the whole world at once. 215 00:11:02.480 --> 00:11:06.240 So the engine aggressively optimizes for high detailed geometry right 216 00:11:06.240 --> 00:11:08.919 in front of the camera. The weapon models the character 217 00:11:09.080 --> 00:11:13.000 arms to ensure the game doesn't waste its precious sixteen 218 00:11:13.039 --> 00:11:16.799 milliseconds calculating light bounces in a room you aren't even in. FPS. 219 00:11:16.840 --> 00:11:21.120 Engines historically rely on spatial partitioning algorithms. 220 00:11:20.480 --> 00:11:23.000 Like binary space partitioning or BSP. 221 00:11:22.759 --> 00:11:27.360 Tree exactly, they mathematically slice the level geometry along planes, 222 00:11:27.759 --> 00:11:31.279 allowing the engine to instantly call or hide any room 223 00:11:31.360 --> 00:11:33.600 that isn't currently intersecting your field of view. 224 00:11:33.960 --> 00:11:37.360 The GPU never even receives the instruction to draw the 225 00:11:37.399 --> 00:11:40.879 monster weighing behind the solid door. It's aggressively wiped from 226 00:11:40.879 --> 00:11:41.799 the rendering cue. 227 00:11:41.840 --> 00:11:45.279 But the moment you change the genre, those specific algorithmic 228 00:11:45.320 --> 00:11:48.960 chets become completely useless. Strip away the walls and drop 229 00:11:49.000 --> 00:11:51.879 the player into a racing simulator going two hundred miles 230 00:11:51.879 --> 00:11:56.039 an hour. A BSP tree fails completely in a sprawling 231 00:11:56.200 --> 00:11:57.240 outdoor environment. 232 00:11:57.440 --> 00:11:59.200 It just can't handle the open space right. 233 00:11:59.480 --> 00:12:03.879 Instead, racing engines divide long corridor tracks into sectors. They 234 00:12:03.879 --> 00:12:08.279 strip out complex AI pathfinding and focus memory bandwidth entirely 235 00:12:08.320 --> 00:12:12.159 on texture streaming. They often utilize old Hollywood tricks like 236 00:12:12.200 --> 00:12:16.039 two D background cards. Ah, yes, those distant mountains aren't 237 00:12:16.080 --> 00:12:18.639 three D models at all, They're just flat images. But 238 00:12:18.720 --> 00:12:21.399 because the camera is hurtling forward so fast, the lack 239 00:12:21.440 --> 00:12:23.159 of parallax is completely hidden. 240 00:12:23.279 --> 00:12:25.159 Then take the camera off the bumper of the car 241 00:12:25.440 --> 00:12:27.519 and suspend it five hundred feet in the air for 242 00:12:27.559 --> 00:12:32.200 a real time strategy game, Suddenly the rendering budget shifts entirely. 243 00:12:32.320 --> 00:12:34.919 You're not looking at one detailed car anymore. 244 00:12:34.559 --> 00:12:37.480 Right, you aren't processing a high fidelity car model. You're 245 00:12:37.639 --> 00:12:42.759 tracking five hundred independent AI units navigating a massive battlefield simultaneously. 246 00:12:43.080 --> 00:12:46.799 Right, and at that scale, running complex collision detection on 247 00:12:47.000 --> 00:12:50.159 every single soldier's footstep would just melt the CPU. 248 00:12:50.320 --> 00:12:54.440 Oh absolutely so, RTS engines just abandon high resolution collision 249 00:12:54.480 --> 00:12:58.080 meshes entirely. They rely on simple grid based height fields 250 00:12:58.320 --> 00:12:59.519 and navigation meshes. 251 00:12:59.600 --> 00:13:02.919 So they just simplify the terrain mathematically exactly. 252 00:13:03.080 --> 00:13:06.600 The engine calculates paths based on a low resolution geometric 253 00:13:06.639 --> 00:13:11.919 abstraction of the terrain, sacrificing visual fidelity for massive computational scale. 254 00:13:11.519 --> 00:13:15.320 And that scale reaches its absolute architectural limit with massively 255 00:13:15.480 --> 00:13:20.039 multiplayer online games because in an MMOG, the true authoritative 256 00:13:20.039 --> 00:13:22.360 engine doesn't even live on the player's hardware. 257 00:13:22.519 --> 00:13:24.639 No, the definitive say of the world resides on a 258 00:13:24.679 --> 00:13:27.480 massive server array in a data center somewhere. 259 00:13:27.600 --> 00:13:30.120 So the local engine running on your machine essentially acts 260 00:13:30.120 --> 00:13:32.240 as a highly advanced dumb terminal. 261 00:13:32.519 --> 00:13:33.720 That's a great way to put it. 262 00:13:33.720 --> 00:13:36.360 It renders the graphics and sends your inputs, but its 263 00:13:36.399 --> 00:13:40.480 primary architectural constraint is masking network latency. It has to 264 00:13:40.559 --> 00:13:44.240 constantly synchronize with the server that is concurrently processing combat map, 265 00:13:44.480 --> 00:13:48.639 inventory databases and micro transactions for ten thousand other clients. 266 00:13:48.879 --> 00:13:51.200 It's juggling a lot, it is, But this brings up 267 00:13:51.200 --> 00:13:53.360 a point I want to push back on. I know 268 00:13:53.440 --> 00:13:56.840 that modern platforms like Unreal in Unity market themselves as 269 00:13:57.039 --> 00:14:02.039 highly modular universal tools. But if the fundamental math required 270 00:14:02.039 --> 00:14:05.039 to call a room in a shooter is completely useless 271 00:14:05.039 --> 00:14:08.159 for pathfinding in a strategy game, why push for a 272 00:14:08.159 --> 00:14:09.279 master engine at all? 273 00:14:09.399 --> 00:14:10.480 That's a very qure question. 274 00:14:10.559 --> 00:14:13.919 I'd assume that attempting to build one master algorithm for 275 00:14:14.120 --> 00:14:18.360 spatial subdivision that dynamically scales across all genres would just 276 00:14:18.399 --> 00:14:23.080 create a bloated, inefficient nightmare. Shouldn't standardizing actually be worse 277 00:14:23.120 --> 00:14:23.919 for performance? 278 00:14:24.480 --> 00:14:29.600 It is the eternal tension in software engineering generality versus optmalody. 279 00:14:29.720 --> 00:14:32.279 You are completely right that a generalized solution carries a 280 00:14:32.320 --> 00:14:33.080 performance text. 281 00:14:33.080 --> 00:14:33.720 So why do it? 282 00:14:33.840 --> 00:14:36.320 Well, if you build a custom racing engine from scratch, 283 00:14:36.879 --> 00:14:39.600 you strip out every line of code related to RTS, 284 00:14:39.879 --> 00:14:43.120 unit pathing or FPS inventory management. You can press your 285 00:14:43.120 --> 00:14:46.639 memory footprint, you maximize your cash coherence, and you squeeze 286 00:14:46.639 --> 00:14:47.759 out a few extra frames per. 287 00:14:47.679 --> 00:14:49.440 Second, Which sounds ideal. 288 00:14:49.480 --> 00:14:53.440 It is computationally, but the financial and temporal cost of 289 00:14:53.480 --> 00:14:58.240 building that bespoke engine is astronomical. A universal engine offers 290 00:14:58.240 --> 00:15:01.919 a sprawling buffet of algorithms. It provides a BSP tree 291 00:15:02.039 --> 00:15:05.080 and a height field navigation system, allowing the developer to 292 00:15:05.159 --> 00:15:06.559 just toggle systems on and off. 293 00:15:06.639 --> 00:15:08.360 Oh I see, so you're trading a little bit of 294 00:15:08.440 --> 00:15:10.080 speed for convenience exactly. 295 00:15:10.120 --> 00:15:13.840 You trade a slight margin of computational optimality for years 296 00:15:13.840 --> 00:15:15.039 of saved engineering time. 297 00:15:15.360 --> 00:15:18.240 Here's where it gets really interesting, though, because the display 298 00:15:18.279 --> 00:15:23.480 paradigms themselves are shifting, and emerging technologies are completely shattering 299 00:15:23.600 --> 00:15:25.159 these traditional trade offs. 300 00:15:24.919 --> 00:15:27.039 Right, the old rules don't always apply anymore. 301 00:15:27.200 --> 00:15:30.799 The rules of genre optimization don't apply when you fundamentally 302 00:15:30.840 --> 00:15:33.919 alter how the brain interfaces with the screen, which is 303 00:15:33.960 --> 00:15:36.039 exactly what happens with virtual reality. 304 00:15:36.320 --> 00:15:40.960 VR forces engine architects to essentially throw out decades of 305 00:15:41.000 --> 00:15:46.320 accepted rendering hacks. The most immediate constraint is stereoscopic rendering. 306 00:15:46.000 --> 00:15:48.480 Which means drawing everything twice right yes. 307 00:15:49.240 --> 00:15:52.159 To create the illusion of depth, the engine must render 308 00:15:52.200 --> 00:15:56.320 the entire scene twice per frame, once for the left eye, 309 00:15:56.600 --> 00:16:01.320 once for the right. From slightly offset virtual care rope positions. 310 00:16:00.919 --> 00:16:03.639 And the frame rate deadlines are unforgiving. We mentioned sixty 311 00:16:03.679 --> 00:16:06.279 frames per second as the standard target for a soft 312 00:16:06.279 --> 00:16:09.440 real time game. In VR, dropping to sixty is a 313 00:16:09.440 --> 00:16:11.279 physiological disaster. 314 00:16:11.039 --> 00:16:14.600 Because of the vestibular system. If a VR engine drops 315 00:16:14.639 --> 00:16:18.320 below ninety frames per second, the latency between the physical 316 00:16:18.399 --> 00:16:20.840 rotation of your head and the visual update on the 317 00:16:20.879 --> 00:16:23.679 screen creates a sensory mismatch in your brain. 318 00:16:23.840 --> 00:16:25.919 The inner ear tells the brain you were moving, but 319 00:16:25.960 --> 00:16:28.320 the eyes report on microscopic. 320 00:16:27.720 --> 00:16:31.440 Delay, and the evolutionary response to that sensory mismatch is 321 00:16:31.480 --> 00:16:35.360 to assume you've been poisoned. Wow really yeah, inducing severe nausea. 322 00:16:35.440 --> 00:16:37.879 So the engine has to push twice as many pixels 323 00:16:38.159 --> 00:16:42.399 significantly faster, with absolutely zero tolerance for dropped frames. 324 00:16:42.559 --> 00:16:46.399 It also breaks standard gameplay mechanics like locomotion. The visual 325 00:16:46.440 --> 00:16:48.840 trick of pushing a joystick forward to slide across a 326 00:16:48.919 --> 00:16:51.200 room works totally fine on a television. 327 00:16:50.840 --> 00:16:55.200 But in VR, smooth artificial locomotion causes instant motion sickness. 328 00:16:55.559 --> 00:16:59.360 Right, So engines had to quickly integrate totally new paradigms 329 00:16:59.440 --> 00:17:04.440 like click teleportation meshes. You even see engines handling vrception 330 00:17:05.079 --> 00:17:07.359 where the player picks up a pair of virtual glasses, 331 00:17:07.720 --> 00:17:11.440 puts them on their virtual head, and seamlessly shifts into 332 00:17:11.480 --> 00:17:15.519 an alternate rendering pipeline within the simulation. It's incredibly meta, 333 00:17:15.640 --> 00:17:19.160 it is, But hardware isn't the only boundary pushing the engine. 334 00:17:19.359 --> 00:17:23.400 The source material highlights the explosion of player authored content. 335 00:17:24.079 --> 00:17:27.240 Look at games like Minecraft or Dreams on the PlayStation. 336 00:17:27.440 --> 00:17:28.839 Tire's doing some wild stuff in. 337 00:17:28.839 --> 00:17:31.920 Those they really are. Players are using in game materials 338 00:17:32.000 --> 00:17:35.599 like Redstone and Minecraft to build fully functioning logic circuits. 339 00:17:35.839 --> 00:17:38.759 If I'm wiring up a complex Redstone logic gait to 340 00:17:38.799 --> 00:17:42.240 automate a massive farm in the game, I'm basically intercepting 341 00:17:42.359 --> 00:17:45.279 and modifying the engine's underlying tick rate. Right. 342 00:17:45.440 --> 00:17:47.720 If we connect this to the bigger picture, you are 343 00:17:47.759 --> 00:17:52.599 engaging in visual programming. Titles like Minecraft and Dreams have 344 00:17:52.680 --> 00:17:55.559 abstracted their engines to such a degree that the game 345 00:17:55.559 --> 00:17:57.599 itself acts as an operating system. 346 00:17:57.720 --> 00:18:00.680 It's almost like you're under the hood without realizing it exactly. 347 00:18:01.000 --> 00:18:04.000 The engine provides a visual interface that allows the player 348 00:18:04.079 --> 00:18:09.240 to manipulate the underlying mathematical simulation, setting boolean variables, writing 349 00:18:09.279 --> 00:18:12.640 conditional logic without ever catching a single line of C 350 00:18:12.880 --> 00:18:16.440 plus plus bag. You are programming the runtime loop using 351 00:18:16.519 --> 00:18:18.240 virtual electrical wire, and. 352 00:18:18.160 --> 00:18:21.279 We are seeing that same level of dynamic simulation replace 353 00:18:21.480 --> 00:18:26.119 traditional art pipelines too. Particularly with character movement, the text 354 00:18:26.160 --> 00:18:30.240 focuses on biomechanical middleware, like a tool called Euphoria. 355 00:18:30.480 --> 00:18:32.240 This is where it gets really sci fi right. 356 00:18:32.359 --> 00:18:36.000 For decades, character movement was basically just a rigid playback 357 00:18:36.039 --> 00:18:38.680 of an animation drawn by an artist. If a character 358 00:18:38.720 --> 00:18:42.000 got shot, they played a pre rendered falling animation. But 359 00:18:42.119 --> 00:18:44.640 Euphoria abandons keyframes for physics. 360 00:18:44.880 --> 00:18:48.640 It utilizes a real time biomechanical model that simulates the 361 00:18:48.720 --> 00:18:52.279 character's muscle mass, their specific weight distribution, and their center 362 00:18:52.319 --> 00:18:52.759 of gravity. 363 00:18:52.799 --> 00:18:54.200 So it's not an animation plane. 364 00:18:54.400 --> 00:18:58.400 No, it employs inverse kinematics and active rag doll physics. 365 00:18:59.279 --> 00:19:01.960 When that digital character takes a hit, there is no 366 00:19:02.119 --> 00:19:07.000 animation file. The physics engine dynamically calculates exactly how their 367 00:19:07.000 --> 00:19:10.119 digital muscles would spasm and how they would stumble to 368 00:19:10.200 --> 00:19:13.640 regain their balance based on the exact vector of the impact. 369 00:19:13.880 --> 00:19:17.400 The complexity required to integrate that level of physical simulation 370 00:19:17.640 --> 00:19:22.799 alongside graphics and audio in just sixteen milliseconds is just staggering, which. 371 00:19:22.599 --> 00:19:26.559 Makes the actual architectural organization of the engine absolutely critical. 372 00:19:26.680 --> 00:19:30.039 I imagine enforcing strict boundaries is the only way to stop, say, 373 00:19:30.160 --> 00:19:33.599 the rendering code from accidentally bleeding into the audio code 374 00:19:33.599 --> 00:19:35.599 and triggering an infinite loop of errors. 375 00:19:35.680 --> 00:19:37.480 Oh, it would be chaos otherwise. 376 00:19:37.720 --> 00:19:41.200 To explain how developers manage this chaos, the text breaks 377 00:19:41.240 --> 00:19:43.079 the architecture down into a layer. 378 00:19:42.880 --> 00:19:47.160 Cake, right, It's an exercise in extreme compartmentalization. At the 379 00:19:47.200 --> 00:19:49.480 absolute bottom of the cake, you have the target hardware 380 00:19:49.519 --> 00:19:52.799 and the operating system. In the early days, a console 381 00:19:52.880 --> 00:19:54.720 was a dedicated games machine. 382 00:19:54.359 --> 00:19:55.119 But not anymore. 383 00:19:55.319 --> 00:19:59.000 No, today, an Xbox or a PlayStation is essentially just 384 00:19:59.079 --> 00:20:03.279 a specialized PE. The engine isn't just talking to the silicon. 385 00:20:03.680 --> 00:20:07.039 It has to negotiate with a hypervisor running background OS 386 00:20:07.039 --> 00:20:10.920 tasks like downloading updates or capturing video buffers. 387 00:20:10.799 --> 00:20:14.119 Which can randomly steal CPU cycles away from the game loop. 388 00:20:14.319 --> 00:20:17.240 Exactly so, to protect the game from those hardware quirks, 389 00:20:17.440 --> 00:20:19.759 you move up a layer to the platform independence layer. 390 00:20:19.960 --> 00:20:22.200 Because a studio obviously doesn't want to write their game 391 00:20:22.200 --> 00:20:24.680 three separate times for Xbox, PC and PlayStation. 392 00:20:24.799 --> 00:20:26.920 That would be incredibly inefficient. 393 00:20:26.559 --> 00:20:30.319 Right, This layer acts like a universal translator using C 394 00:20:30.480 --> 00:20:34.279 plus plus wrapper classes. The core game logic just says 395 00:20:34.559 --> 00:20:38.680 allocate this memory or read this file. The rapper function 396 00:20:38.799 --> 00:20:42.000 intercepts that generic command and automatically translates it into the 397 00:20:42.079 --> 00:20:46.240 highly specific proprietary system call required by whatever hardware it 398 00:20:46.279 --> 00:20:47.200 happens to be running on. 399 00:20:47.319 --> 00:20:49.720 And sitting above that hardware abstraction, you find the third 400 00:20:49.720 --> 00:20:54.720 party stk's software development kits. Very huge studios build every 401 00:20:54.720 --> 00:20:56.079