Model architecture
Defines the mathematical operations, representations, and learning mechanisms a model uses.
Seven Layers · Layer 6
AI hardware is the physical foundation of artificial intelligence. It includes processors, memory, networking, storage, sensors, packaging, power systems, cooling equipment, and the manufacturing capacity needed to turn a model design into a functioning machine.
Software determines what an AI system can attempt. Hardware determines how quickly, cheaply, reliably, and widely it can do it.
Layer 6 is broader than the semiconductor chip alone. A powerful accelerator is useful only when it can receive data from memory, communicate with other processors, remain within power and temperature limits, and be manufactured in sufficient volume. Modern AI performance therefore emerges from a complete physical system rather than from a single component.
Defines the mathematical operations, representations, and learning mechanisms a model uses.
Schedules, deploys, serves, evaluates, and operates models across computing resources.
Provides processors, memory, interconnects, storage, power, cooling, sensors, and physical machines.
Uses the capabilities of the layers below to discover new medicines, materials, theories, and designs.
AI workloads demand enormous numbers of mathematical operations and move equally enormous quantities of data. The challenge is not simply to add more transistors. The system must balance compute throughput, memory bandwidth, communication speed, energy use, heat, manufacturing yield, reliability, and cost.