The Graphics Processing Unit is a massively parallel processor that excels at performing the same operation on thousands of data points simultaneously. Originally designed exclusively for 3D rendering, the GPU has become the primary accelerator for artificial intelligence, scientific simulation, video encoding, and cryptocurrency mining. The fundamental architectural difference between CPU and GPU reflects their respective design goals. A CPU is optimized for sequential, low-latency execution of complex, branchy code — it dedicates most of its die area to branch prediction, out-of-order execution logic, and large caches. A modern CPU has 8–24 powerful cores. A GPU is optimized for throughput on regular, predictable workloads — it dedicates most of its die area to thousands of simple arithmetic units (CUDA cores, Stream Processors) arranged in a hierarchy: cores → Warps/Waves → Compute Units → GPCs. The 3D rendering pipeline begins in the application (game or engine) which generates draw calls containing geometry (triangles defined by vertex positions) and shader programs. The GPU vertex shader transforms 3D world-space coordinates into 2D screen-space coordinates for each vertex. The rasterizer converts the 2D triangles into fragments (candidate pixels). The pixel shader calculates the final color of each fragment based on lighting, textures, and material properties. The output merger combines overlapping fragments (depth testing, alpha blending) and writes the final image to the VRAM framebuffer. This framebuffer is then scanned out to the display via HDMI or DisplayPort at the refresh rate. Ray tracing replaces or supplements rasterization's approximations of lighting with physically accurate light simulation. Each ray is traced from the virtual camera through each pixel, bouncing off surfaces and accumulating lighting contributions. This accurately simulates reflections, refractions, global illumination, and ambient occlusion. The computational cost is 10–100× higher than rasterization for equivalent image quality, making dedicated RT cores (NVIDIA) and ray accelerators (AMD) essential for playable framerates. VRAM (Video RAM) is the GPU's dedicated memory pool. All textures, framebuffers, shader programs, and geometry data must fit in VRAM for optimal performance. When VRAM is exhausted, the GPU begins using system RAM over the PCIe bus — bandwidth drops from 1000+ GB/s (GDDR6X) to 64 GB/s (PCIe 5.0 x16), causing massive stuttering. 4K gaming with high-resolution texture packs easily requires 12–16GB VRAM. AI image generation models (Stable Diffusion, ComfyUI) require 8–24GB depending on model size.