Discrete GPU

A dedicated graphics processing unit separate from the CPU, designed for parallel processing tasks such as rendering, scientific computing, and Artificial Intelligence inference.

Architecture & Memory Models

  • Traditional Discrete GPUs: Utilize vram (Video RAM) physically separate from system ram. Data transfer occurs via the PCIe bus, creating bandwidth bottlenecks for large models.
  • Unified Memory Architectures: Emerging platforms allow the GPU to access system memory directly, eliminating the need to duplicate data between CPU and GPU memory spaces. This is critical for running large language models (LLMs) that exceed traditional VRAM capacities.

Emerging Platforms: AMD Ryzen AI Halo

The AMD Ryzen AI Halo developer platform represents a shift toward unified memory systems for local AI development, challenging the traditional discrete GPU model for specific workloads.

Comparison with Integrated Graphics

FeatureDiscrete GPUIntegrated GPU (iGPU)Unified Memory (e.g., Ryzen AI Halo)
MemoryDedicated VRAMShared System RAMUnified System/VRAM Pool
PerformanceHigh (Parallel Compute)Low to ModerateHigh (for AI/ML workloads)
FlexibilityFixed VRAM capacityLimited by system RAMScalable with system RAM
Use CaseGaming, 3D Rendering, Heavy ComputeOffice tasks, light mediaLocal LLMs, AI Development

See Also