PPU

A physics processing unit is a specialized microprocessor that offloads physics calculations from the main CPU. It handles the numerical work of collision detection, rigid body dynamics, fluid dynamics, and constraint solving, returning results that control how objects move and interact in real time. By dedicating silicon to these mathematically intensive tasks, a PPU frees the general-purpose processor to manage rendering, input, logic, and other workloads.

The most widely known PPU was NVIDIA's PhysX, released in the mid-2000s as a discrete PCI card. It performed matrix operations and iterative solvers for rigid body systems, soft body dynamics, and particle effects at speeds that outpaced CPU-only implementations by significant margins. The architecture relied on parallel computation: thousands of small collision tests, force calculations, and constraint updates could execute simultaneously across the chip's processing elements.

Dedicated PPU cards became rare as CPU cores multiplied and GPUs grew more programmable. Physics calculations now typically run either on multi-core CPUs using vectorized instructions (SSE, AVX) or on GPUs through general compute frameworks. Some industrial simulation software still uses the term PPU to describe physics acceleration hardware or physics-aware co-processors, though the discrete card product category has largely vanished.

The main technical trade-off remains simple: physics calculations demand high throughput (many small operations) and high latency tolerance (results can be computed a frame or two late), while rendering demands low latency and complex state management. A dedicated PPU was well-suited to the first task; modern multi-threaded CPUs and compute shaders handle both adequately, making the specialized die less economical than before.