Stratum: System-Hardware Co-Design with 3D-Stackable DRAM for Efficient Moe

For those confused by the headline, HN changed it from MoE to Moe.

Though, thinking about "efficient moe" I'm fondly reminded of projects like https://make.girls.moe/ (2017) - "Towards the Automatic Anime Characters Creation with Generative Adversarial Networks" [sic] https://arxiv.org/abs/1708.05509

A simpler time, for sure.

Seems to me similar hardware with 3D-stackable RAM could be very useful also for more "traditional" workloads.