Google's "Frozen v2" chip hardcodes Gemini architecture into silicon
The hardware targets tenfold TPU efficiency gains by 2028.
Combined views
166.1K
11 Sources, first seen ago
Sources
CH 
Chubby♨️@kimmonismusGoogle may be preparing to freeze parts of Gemini’s architecture directly into silicon. Informally called "Frozen v2," the chip reportedly targets 6–10× more tokens per watt than Google’s newest TPUs. Deployment is planned for as early as 2028. The motivation is immediate:…
- likes: 722
- replies: 55
- bookmarks: 110
- reposts: 50
XJ xjdr@_xjdrLove my Google friends but maybe make a good model first and _then_ focus on inference efficiency after people actually want to use it ? https://twitter.com/AndrewCurran_/status/2079220150168138204
- likes: 196
- replies: 21
- bookmarks: 14
- reposts: 3
AC Andrew Curran@AndrewCurran_Google is developing a new chip named 'Frozen v2' specifically designed to run the Gemini family of models more efficiently. The new hardware is expected to arrive in 2028. This was originally reported by The Information. Gemini probably helped design this.
- likes: 362
- replies: 30
- bookmarks: 26
- reposts: 27
AE Amir Efrati@amir🔥chip news: Google planning to bake LLM architecture into new AI server chips (non-TPUs) to run Gemini 6-10x more efficiently
- likes: 11
- replies: 4
- bookmarks: 4
- reposts: 2
PD Pedro Domingos@pmddomingosIt’ll be obsolete by the time it’s in production. https://twitter.com/ns123abc/status/2079217148669628482
- likes: 210
- replies: 17
- bookmarks: 24
- reposts: 7
BT Bojan Tunguz@tunguzSounds like ASICs with extra steps. https://twitter.com/mtslive/status/2079198478849413390
- likes: 38
- replies: 3
- bookmarks: 5
- reposts: 2