Post
GPUs are not just for drawing triangles anymore -- they are massively parallel supercomputers sitting in your PC waiting to be used.
GPU compute uses the graphics card's thousands of parallel processing cores for general-purpose calculations that are not directly related to rendering. Compute shaders run arbitrary code on the GPU, and because GPUs can execute thousands of threads simultaneously, they excel at data-parallel problems: particle simulations with millions of particles, fluid dynamics, cloth physics, terrain deformation, AI inference, pathfinding for large groups, and physics simulations. The programming model is different from CPU code -- you write kernels that operate on large arrays of data in parallel, and the GPU's scheduler distributes work across its cores. DirectCompute, OpenCL, Vulkan Compute, and Metal Compute Shaders are the common APIs. The results can be dramatic -- tasks that take seconds on a CPU can complete in milliseconds on a GPU.
Example
Teardown uses GPU compute for its voxel-based destruction physics, simulating millions of voxels being fractured, launched, and settled by physics in real time. The entire game world is destructible because the GPU handles the massive parallel computation that would be impossible on a CPU alone.
Why it matters
GPU compute unlocks simulation scales that are simply impossible on CPUs. As games push toward larger worlds, more particles, more physics objects, and more sophisticated AI, the GPU becomes essential compute infrastructure rather than just a rendering device. The boundary between graphics and general computation is dissolving.
Related concepts