▶
abstract
Modern video games must compute complex simulations and render detailed 3D scenes within 16.6 milliseconds per frame. This talk explores how parallel computing has shaped game development — from the single-core Game Boy to the 2,304 shader cores of a PlayStation 5. Tracing the exponential growth of console hardware over four decades, it examines how the GPU's embarrassingly parallel architecture enables real-time 3D rendering, illustrated with GPU traces from a shipped title. In contrast, gameplay logic resists parallelization due to scattered memory access, complex inter-object dependencies, and the need for deterministic execution order. The talk closes by comparing two engine architectures that address this tension — Unreal Engine's dedicated-thread pipeline and id Tech 7's fully job-based approach — and argues that the choice between them is ultimately a game design decision.