AI summary by TwinMind
Tim Sweeney: Fortnite, Unreal Engine, and the Future of Gaming | Lex Fridman Podcast #467
Lex Fridman
4h 25m 19s
Computer Graphics and Human Rendering Challenges
- Humans hardest part of computer graphics due to millions of years of evolution
- Dedicated brain systems to detect patterns, faces, infer emotions/intent
- Cavemen needed to determine if strangers were friendly or threatening
- Extraordinarily detailed expectations of faces, notice imperfections from CG limitations
- Advanced hardware for capturing humans
- Capture sphere with dozens of cameras taking high resolution, high frame rate video
- Range of motions, facial muscle/tendon/fat interactions for expressions
- Not just shape but entire range of motion possibilities
- Hair rendering extremely difficult
- Can't render every hair (billion times too slow)
- Need approximations capturing net effect without calculating every light interaction
- Subsurface scattering critical for realistic skin
- Skin not completely opaque, light travels through it
- Huge impact on appearance
- Why painted mannequins never look realistic
- Oiliness creates specific nuanced complex reflections
- Light absorbed and travels through skin creating textures human eyes perceive
- Subtle expressions: squinting, smile formation, difference between real/fake smiles
- Getting faces right requires dozens of different systems interplaying
- If any one wrong, eye completely drawn to it (uncanny valley)
Tim Sweeney's Programming Origins and Early Epic Games
- First programming experience age 11 visiting brother Steve (16 years older) in California
- Brother had IBM PC, learned BASIC in few days
- Vivid memory of first programs: asterisk-catching game in text mode
- 10-15,000 hours coding age 10-20 before shipping first commercial program
- Built databases, programming language/compiler (didn't know where to buy one)
- Bulletin board systems before internet
- Used modem to dial local phone numbers
- Every town had hundreds of bulletin boards with different themes
- ZZT origin story (1991)
- Started as text editor project
- Added smiley face cursor, different characters performed gameplay actions
- Evolved into game editor with ~20-30 objects
- User testing with neighbors of all ages
- Released first game free, sold sequels for $30 via mail order
- Making $100/day as 20-year-old
- Shareware model precedent to free-to-play
- Always included editor tools with games - core Epic philosophy
Unreal Engine Development Journey
- Wolfenstein/Doom breakthrough moment - future was 3D
- Initially depressed by Carmack's technical wizardry, stopped programming 6 months
- Michael Abrash's articles explained 3D texture mapping techniques
- 3.5 years development, constantly thinking 6 months from shipping
- Team grew to ~20 people, nearly bankrupted Epic multiple times
- Early engine licensing deals (Microprose, GT Interactive) provided crucial funding
- Technical innovations:
- Constructive solid geometry in real-time (30-hour coding session)
- Dynamic colored lighting with shadows
- Volumetric fog (inspired by Finnish GPU company screenshot)
- Light mapping techniques for performance
Unreal Engine Evolution (Versions 1-5)
- UE1: Software rendering, later Voodoo GPU support
- UE2: GPU hardware acceleration, vehicles
- UE3: DirectX9 programmable shaders, real-time lighting, multi-threading
- UE4: Continued improvements, more artist options
- UE5: Major leap with Nanite (micropolygon geometry) and Lumen (global illumination)
- Some systems unchanged since 1998 (file management, networking)
- Single-threaded game simulation major limitation (using 1 core of 16-core CPU)
- UE6 plans: Address core limitations, multi-threading, few years away
Nanite and Lumen Technical Details
- Nanite: Virtualized micropolygon geometry system
- Handles billions of polygons efficiently
- Nyquist sampling theorem: 2 triangles per pixel indistinguishable from thousands
- Bypasses GPU triangle rasterization hardware for small triangles
- Uses pixel shaders and ray tracing for better results
- Lumen: Global illumination system
- Calculates light bouncing throughout entire scene
- Multiple scales from kilometers to millimeters
- Seamless integration across all levels
- Dynamic lighting that "just works" for artists
Metahuman Technology
- Vlad Mastilovic leading digital humans effort
- Capture sphere hardware with dozens of cameras
- Thousands of hours processing per human capture
- Dataset goal: encompass entire range of human faces across all cultures/ages
- Ability to reconstruct arbitrary faces from iPhone photos using learned data
- Metahuman Creator: Tool for creating faces with adjustable parameters
- Metahuman Animator: Facial capture from devices like iPhone
- Translation between actor face shape and character face shape
Fortnite Development and Success
- Origin: 2011 game jam, build forts by day, defend against zombies by night
- 7 years development, multiple pivots
- Art style change from realistic to Pixar-inspired stylized
- Save the World mode launched early 2017 (moderate success)
- Battle Royale: 4-week development in war room with 30 people
- Explosive growth: 300 to thousands of employees, hundreds of millions to billions revenue
- Technical scaling: 40,000 to 15 million concurrent users
- Microservices architecture, AWS cloud hosting crucial
- Cross-platform breakthrough with Sony (initially resistant, eventually opened up)
Verse Programming Language
- Functional logic language for large-scale simulation programming
- Zero, one, or multiple values from expressions (failure, success, iteration)
- Eliminates boolean variables in favor of success/failure patterns
- SQL-like queries in conditions and loops
- Type system enables mathematical proofs (Curry-Howard correspondence)
- Backwards compatibility checking through type system
- Transactional memory for concurrency
- Speculative execution of transactions
- Conflict detection and retry for conflicting operations
- Moving from programmer problem to language problem
- Currently shipping in UEFn (Unreal Editor for Fortnite)
- Will be central to UE6
Metaverse Vision and Standards
- Current problems: Fragmented ecosystems
- Different friends systems across platforms
- Different usernames, controls, purchases don't transfer
- No social cohesion between games
- Solutions needed:
- Federated social systems with @ usernames
- Unified economies for cosmetic items across games
- Standards bodies for portable outfits/items
- Revenue sharing based on engagement patterns
- Programming language standards crucial (like HTML/JavaScript for web)
- Need for distributed collaborative content creation
Epic vs Apple Legal Battle
- Apple's 30% fees economically ruinous, more than most game companies' profits
- Gatekeeping prevents competition, innovation
- Epic uniquely positioned to fight: majority users on PC/console, could survive iOS loss
- Apple's soft power: slowing updates, app store manipulation
- Other companies terrified of Apple destroying their business
- Apple's anti-competitive techniques:
- Blocking web browser improvements on iOS
- Preventing competing stores from integrating social features
- Complex barriers even where legally required (EU DMA)
- Philosophical difference: Apple II was completely open platform vs current closed ecosystem
Epic Games Store and Competition
- 12% revenue cut vs Steam's 30%
- Launcher acknowledged as clunky, needs improvement
- Exclusive deals as competitive necessity when price competition blocked
- Steam's soft power prevents developers from offering better prices on Epic
- Google's "Project Hug" paid hundreds of millions to prevent exclusive deals
- All-in cost of operating stores ~6% of revenue (Google Play data)
- Goal: Cross-platform social features, unified friend systems
Future of Gaming Industry
- Trend toward large-scale multiplayer social experiences
- Metcalfe's Law: Value quadratic with percentage of real-world friends connected
- Consolidation into fewer, bigger games with massive reinvestment
- Two main opportunities:
- Big multiplayer experiences that retain players long-term
- Efficient single-player games as "vacations" from main games
- Winner-take-all dynamics emerging
- Fun remains the core principle - industry loses sight of this at its peril
