Mythic 的模拟存内计算架构
Mythic's analog compute-in-memory architecture

原始链接: https://www.mythic.ai

传统芯片每秒在处理器和内存之间传输数据数十亿次,大部分能量都消耗在数据搬运而非计算上。这是一个 80 年来从未被行业质疑过的架构缺陷。Mythic 做到了。我们将人工智能模型权重存储在闪存内,并在源头直接进行模拟计算,从根源上消除了这种浪费。

抱歉。
相关文章

原文

Traditional chips shuttle data between processor and memory billions of times per second, burning most of their energy on movement, not computation. It's an 80-year-old architectural flaw the industry never stopped to question. Mythic did. We store AI model weights inside flash memory and compute in analog, directly at the source, eliminating the waste at its origin.

联系我们 contact @ memedata.com