IGP

An integrated graphics processor is a graphics processing unit fabricated on the same silicon die as the central processor, eliminating the need for a separate discrete graphics card. The IGP shares the system's main memory (typically via a portion of RAM allocated as video memory) rather than maintaining its own dedicated VRAM, which reduces component count, power consumption, and the physical footprint of the device.

IGPs are standard in consumer laptops, compact industrial PCs, and embedded systems where graphics output is needed but discrete GPUs would be impractical. They handle 2D rendering, basic 3D acceleration, and video decoding at modest frame rates. Modern IGPs can process light CAD work, video playback, and machine vision preprocessing tasks. However, their memory bandwidth is limited by the CPU-to-RAM interface, and they have fewer execution units than dedicated cards, making them unsuitable for real-time 3D simulation, scientific computing, or high-throughput data parallel work.

Performance and memory architecture

The GPU portion typically occupies 10 to 30 percent of the die area. Its performance scales with clock speed and core count, but bottlenecks appear when system memory bandwidth becomes saturated. In industrial settings, an IGP's inability to buffer large datasets independently means applications must manage memory transfers carefully. Thermal design power (TDP) is shared between CPU and GPU, constraining peak performance if the processor is already warm.

IGPs have become substantially more capable over the past decade. Entry-level laptop and desktop processors now include IGPs with hardware video decode for H.264 and HEVC, reducing CPU load during streaming and archival applications. In factory automation and machine vision systems, this decode capability offloads video processing from the main processor core, freeing cycles for logic and control tasks.

The trade-off between cost, power, and capability makes IGPs dominant in industrial control systems that require graphical HMI output without the expense of a discrete card. Their weakness in parallel data processing means tasks like image filtering, neural network inference, or signal processing should still run on dedicated accelerators or the CPU itself, not pushed to the IGP expecting a speedup.