[KongchangAI]
Concept统一内存架构 / Unified Memory Architecture

UMA

An architecture in which the CPU and GPU share the same physical memory. It removes data copy overhead and benefits workloads such as AI inference that move data frequently.

Timeline (last 90 days)

Oct 1

EP Buffer supports the FP8 data type and Cache Handle-related operations, and uses UMA to advance communication in the background while freeing AIV cores

Unverified50%

All Facts (1)

Source Articles