@kimmonismus I wish it had more ram bandwidth. (Maybe wider bus?) For llm inferring, I feel like thatโs the only bottleneck of this and dgx spark. Tons of unified memory, Iโm glad.
@SawyerMerritt i don't think people like that realize iot use cases on top of business/consuper isp. eventually, any edge devices will use starlink or satellite connection at the end.
we may even see what elon claims, satellite shooting out tokens, edge devices doing inference.