CPU不是不能做ai运算,只不过效率低一些,但通用性强,,可编程性最好。现在CPU甚至能模拟显卡工作,但反过来gpu基本不可能模拟CPU全部工作。gpu要编程还只有唯一cuda工具,还是封闭的工具,看看CPU多少编程工具,低级汇编,高级c c++ 等等一大堆编程工具,gpu一开始出现是为多边形3D加速开发的,并行计算更好,每个计算单元都不大,误打误撞发现适合ai运算,ai核心算法就是多维空间向量变换和计算,这个3d空间多边形图形计算很相似。
Whatever it's IPU or GPU component, the HPC unit must be manufactured by TSMC. We will upgrade this territory of AI very quickly in the coming future.
@flarelu3157 Жыл бұрын
IT'S IMPOSSIBLE, Americans will always come for you if you are capable of stealing any piece of bread from them. Nonetheless, if you keep serving them like a dog, you will be petted with care and love.
@georgechang4163 Жыл бұрын
@@flarelu3157 If you served them like a shit, you can say to serve PRC becomes worse motherfucker than to serve USA. Even to the last moment, you don't get any sympathy from PRC. The proverb said never to trust a born liar.
@georgechang41633 ай бұрын
One year was past so fast. Do you know what kind of situation now?
@@terryjoun1687 而且主要幾個科技大廠,例如GOOG, AAPL, TSLA, BABA 等都卯起來自己搞NPU 或者IPU 之類的。就會出現的就算不是NVDA ,也會是這些大廠之一
@eruisiwang11 ай бұрын
这个IPU和compute in memory什么差别?
@jasenliu10 ай бұрын
dream is always sweet and reality is to catch up the leader is always hard, the best chance is to wait leader make a mistake, a series mistakes, because chasers are more likely to make mistake than leader.