EpochLex entry
GPU Offloading
G-P-U OFF-loh-ding
A model-execution technique that distributes model weights, caches, or computation between GPU memory and slower memory tiers such as CPU RAM or disk so workloads can run within limited GPU memory.
EpochLex entry
G-P-U OFF-loh-ding
A model-execution technique that distributes model weights, caches, or computation between GPU memory and slower memory tiers such as CPU RAM or disk so workloads can run within limited GPU memory.