AI 용어집
인공지능 완전 사전
Edge GPU Computing
Distributed computing architecture leveraging the capabilities of embedded GPUs at the network edge to execute AI workloads locally, thereby reducing latency and dependency on the cloud.
NVIDIA Jetson
NVIDIA's embedded computing platform integrating optimized GPU and CPU for edge AI, offering an all-in-one solution for deploying ML models under low power and small size constraints.
Inference Acceleration
Hardware and software optimization aimed at reducing ML model prediction time through GPU parallelization, specialized compilation, and reduction of redundant operations.
CUDA Core
Fundamental computing unit in NVIDIA GPUs enabling parallel execution of thousands of threads simultaneously, essential for accelerating matrix operations in edge deep learning.
Embedded GPU
Graphics processor integrated into embedded systems with specific power, size, and thermal constraints, optimized for AI workloads in edge environments.
Power Efficiency
Optimized performance/energy consumption ratio for edge devices, requiring algorithms and hardware architectures designed to maximize operations per watt consumed.
GPU Memory Management
Optimization of data allocation and transfer between RAM and VRAM on embedded systems, including pooling techniques, batch size adaptation, and memory pinning for maximum performance.
Pruning Technique
Method of selectively pruning unnecessary neural connections in deep learning networks, reducing computational and memory complexity for efficient deployment on edge GPUs.