5 ms·Imagine having machine learning model weights just be encoded in blocks of SRAM right on the CPU die.by ummonk 4y agoImagine having machine learning model weights just be encoded in blocks of SRAM right on the CPU die.alberth 4y agoThat already exists. https://youtu.be/GVsUOuSjvcg https://youtu.be/GVsUOuSjvcgastrange 4y agoConversely, you can start putting ALUs and matrix multiplies in the DRAM.sidpatil 4y agoThis is what I thought they meant when they mentioned "in-memory compute".