Infrastructure
aka Graphics Processing Unit
GPU
Massively-parallel accelerator that dominates neural-network training and inference.
Definition
GPUs provide thousands of cores and high-bandwidth memory ideal for the matrix multiplications behind deep learning. NVIDIA dominates AI GPUs (H100, B200, GB200); AMD's MI300 and AWS Trainium/Inferentia provide competitive alternatives.
Common use cases
- Training
- Inference
- Research