MemryX, a know-how start-up originating from the College of Michigan, has launched a $149 M.2 module [PDF] designed to supply environment friendly AI processing capabilities for compact computing programs. The module targets edge computing purposes, the place energy effectivity and compact design are important.
Packing 4 MemryX MX3 AI accelerator chips onto an ordinary M.2 2280 kind issue, allows the module to be simply built-in into programs outfitted with a PCIe Gen 3 M.2 slot. Every MX3 chip is alleged to ship 6 TOPS (Tera Operations Per Second), providing a complete of as much as 24 TOPS of compute energy whereas consuming simply 6 to eight watts of energy. It additionally helps a variety of knowledge codecs, together with 4-bit, 8-bit, 16-bit weights, and BFloat16. Notably, the module operates with out lively cooling, counting on an included passive heatsink to handle thermal efficiency.
Phoronix has examined the MemryX MX3 M.2 module, evaluating its efficiency and usefulness as an AI accelerator. Put in in a system with an M.2 PCIe Gen 3 slot working Ubuntu 24.04 LTS, the module proved easy to combine because of MemryX’s open-source drivers and developer instruments. The module’s 24 TOPS efficiency from its 4 MemryX MX3 chips is alleged to be adequate for varied inference workloads, significantly these optimized for 8-bit weights.
Picture 1 of 3
