Google chuan Cloud Next conference-ah an AI chip thar ber **Ironwood** chu an puang a. An TPU (Tensor Processing Unit) version pasarihna a ni a, inference atana ruahman bik a hmasa ber a ni bawk,chu chu AI model kalpui tur atana siam a ni tihna a ni a, training mai ni lovin.
Ironwood-a thil thar awmte chu hetiang hi a ni:
- Size pahnih a awm a: pakhatah hian 256 chips a awm a, a lian zawk chu 9,216 chips a ni.
- Chip pakhat hian Second khatah operation trillion 4,614 (TFLOPs) thleng a thleng thei.
-192GB fast memory a nei a, super high-speed bandwidth(second khatah 7.4 terabits) a nei bawk.
- A chhunga core bik SparseCore tih hian recommendation systems ang chi hna (e.g. eng thawmhnaw nge lei tur tih rawtna) ah a pui thin.
-rang, energy hmang tlem, leh data sawn kual vela delay tihtlem theih tura duan a ni.
- Ironwood hi Google Cloud’s AI Hypercomputer-ah hman a ni dawn a, hei hi AI hnathawhna lian tham tak atana setup chak tak a ni.
Google chuan he chip hi an chip a chak ber leh efficient ber a nih thu an sawi a, a hma aia tha zawka AI hna lian tak tak, real-time handle thei tura siam a ni.