GPU Driven Rendering (GPU-driven rendering / GPU Driven) is a technical term that is extremely important in the rendering and optimization of game graphics (technical art). An explanation of a next-generation rendering architecture that completely transfers object culling (view frustum culling">view frustum culling, occlusion) and drawing command generation to the GPU's Compute Shader to reduce the draw call load on the CPU to zero.

An example in the real world: DX of a delivery factory where the president (CPU) of the head office completely stops checking tens of thousands of deliveries (3D models) one by one at the site and manually issuing delivery slips (draw calls) every morning, and sends all bundles of delivery slips to the on-site ultra-high-speed robot (GPU) and autonomously dispatches them on-site

GPU-driven rendering is a process that has previously been performed by the CPU around the clock. href="/article/term-draw-call">draw callIssuance)'' work is entirely transferred to the GPU itself, a next-generation DX (automation) project for rendering factories.'' The CPU simply throws the ``positional data list of objects here'' to the GPU's shared buffer once when the game starts. After that, the GPU automatically issues draw calls on its own to draw only those who are on the screen in a massively parallel manner (Indirect Drawing), so the CPU's draw call-related load is almost completely reduced to zero, and the information density of the entire screen can be dramatically increased by more than 10 times compared to conventional methods.

GPU Driven Rendering (GPU Driven Rendering / GPU Driven) concept infographic diagram

Illustration: GPU Driven Rendering (GPU Driven Rendering / GPU Infographic that illustrates the basic processing flow and mechanism of Driven) in Japanese in an easy-to-understand manner

Detailed mechanism and operating principle

In Unity 6 or later, enable the "GPU Driven Rendering" function such as "GPU Resident Drawer", upload the object position list to the GPU as a Structured Buffer, and complete everything from culling to generation of drawing instructions (DrawProceduralIndirect) within the GPU's massively parallel Compute Shader.