Xiaomi has just put together an AI Cube development machine—packing three of its own Xuanjie chips into a mini PC, with a focus on local large-model inference.
The specs are pretty impressive:
A 120B model can run, and it also supports one-click switching to a 3B fast model. Bandwidth goes straight to 1.22TB/s, and it can reach up to the 200B-class at maximum.
This wave of local AI and edge computing is really accelerating. Once the hardware gets cranked up, the race tracks related to compute and model capabilities are likely to get even more lively.
Has anyone been watching the AI sector lately? Do you think local inference could spark another round of sentiment? $HK1810 #小米
The specs are pretty impressive:
A 120B model can run, and it also supports one-click switching to a 3B fast model. Bandwidth goes straight to 1.22TB/s, and it can reach up to the 200B-class at maximum.
This wave of local AI and edge computing is really accelerating. Once the hardware gets cranked up, the race tracks related to compute and model capabilities are likely to get even more lively.
Has anyone been watching the AI sector lately? Do you think local inference could spark another round of sentiment? $HK1810 #小米
