gpu

Definition

Originally a graphics chip, now the default substrate for deep learning because its thousands of parallel lanes map cleanly onto dense matrix multiplication. Training and inference costs are quoted in GPU-hours, and on-board memory is usually the binding constraint on how large a model you can hold. The advantage over a CPU is not clock speed but arithmetic throughput and memory bandwidth against highly regular work.

Explainers using this term

• Paper Explained: Agentic ESOpt — Drop Backprop, Jiggle the Weights, and Train Long-Horizon LLM AgentsAgentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirements