We have been quietly building something that attacks one of AI’s biggest unsolved problems: the memory wall.
Constantly moving massive model weights between memory and compute is the real driver of high inference cost and power consumption.
We think there is a better way.
🧵 1/8
7/8 Archality was founded by CPU and GPU industry veterans. Our goal is practical, deployable technology that compresses inference costs while working within real-world data center constraints.
@madmaxbr5@xennygrimmato_ There are a couple of people proposing this with prototypes currently. Non digital domains and non Von-Neumann architectures are a difficult sell to VC's when their technical advisors only know what they were taught (many never practiced); even if you have the ability to make it.
@CRudinschi@xennygrimmato_ There's lots of systems design that's hard to tackle at this point. A lot of licensing issues patent wise are going to surface with PIM; you'll see filings emerge that cover the entirety of the ideas people are taking to production soon.