dGPU
Initialism of discrete graphics processing unit, a graphics processing unit which has access to dedicated graphics memory.
dGPU: a dedicated graphics chip with its own memory
A discrete GPU is a graphics processing unit installed as a separate component, not integrated into the CPU. It has its own dedicated memory, typically GDDR6 or GDDR6X VRAM ranging from 2GB to 48GB depending on the model, which the processor accesses without competing for system RAM. This separation allows the GPU to operate independently and perform rendering, compute, or AI tasks in parallel with the CPU.
The term "discrete" distinguishes these cards from integrated graphics, where GPU cores and memory sit on the same die as the processor. A dGPU connects to the motherboard via PCIe, usually PCIe 4.0 or 5.0 in current systems. Common manufacturers include NVIDIA (GeForce, RTX, Tesla lines) and AMD (Radeon, MI series). Older systems used AGP slots; some industrial or embedded applications still use PCI or proprietary connections.
In industrial settings, dGPUs serve different purposes than consumer gaming. Data centers deploy them for parallel compute tasks: machine learning inference, scientific simulation, video transcoding, and data analytics. Manufacturing facilities use them for real-time computer vision in quality control systems. Mining operations historically relied on them for cryptocurrency hashing. They also appear in workstations for CAD rendering, 3D modeling, and digital content creation where throughput matters more than latency.
Thermal and power considerations
A discrete GPU generates significant heat. Most modern dGPUs require 75 to 450 watts of power, supplied through 6-pin, 8-pin, or dual 8-pin connectors. The card's cooling solution (fan, heatsink, or liquid block) must handle continuous load. Industrial environments often run dGPUs in quiet, thermally controlled cabinets with airflow management. Passive cooling is rare for anything more than light workloads.
The distinction from integrated graphics matters for procurement and system design. A discrete GPU adds cost, power consumption, and complexity but delivers higher throughput and isolation of workloads. In industrial systems where reliability and determinism are critical, this isolation is a benefit: GPU failure does not crash the main processor, and GPU memory can be managed separately from system heap. However, poor PCIe driver support or firmware mismatches can introduce latency and data corruption, especially in real-time environments.