谷歌被曝要把Gemini“写进”芯片？推理能效最高提升10倍
Quick Answer
Google is developing the Frozen v2 chip, integrating parts of the Gemini model to enhance AI inference efficiency by 6 to 10 times.
Quick Take
This specialized chip aims to address AI computing shortages and is expected to be deployed by 2028.
Key Points
- Frozen v2 aims to reduce computational decisions and data movement during AI inference.
- Google's engineers estimate a 6 to 10 times efficiency increase compared to existing AI chips.
- The chip is part of Google's strategy to alleviate internal AI computing shortages.
- Frozen v2 is not intended to replace TPU but serves as a specialized solution for Gemini.
- Deployment of Frozen v2 is projected for 2028, pending model architecture stability.
DeepSignal Analysis
What happened
Google is developing a new AI server chip called Frozen v2, which aims to integrate parts of the Gemini model to enhance AI inference efficiency by 6 to 10 times. This chip is intended to address AI computing shortages and is expected to be deployed by 2028.
Key evidence
- The Frozen v2 chip will integrate parts of the Gemini model to improve AI inference efficiency, potentially achieving 6 to 10 times the output per unit of power compared to Google's latest AI chips.
- Google's internal conflicts over AI computing shortages have led to the refusal of some external customer orders, highlighting the urgency of the Frozen v2 project.
- The initial version of the Frozen chip was shelved due to concerns over its limited lifespan and adaptability, leading to the current approach of not fully embedding model weights into the chip.
Why it matters
The development of Frozen v2 reflects Google's strategy to enhance AI performance while addressing significant computing shortages. By integrating specific model architectures directly into hardware, Google aims to reduce operational costs and improve efficiency. However, this approach raises concerns about flexibility, as future updates to the Gemini model may not be compatible with the chip.
Source Excerpt
谷歌拟于2028年部署Frozen v2，固化Gemini部分架构，推理能效或提升6至10倍。
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from WebSearch (Tavily)
See more →lila ayu
The 8-week hands-on track focuses on mastering Generative AI, , QLoRA fine-tuning, and AI Agents, enabling participants to build 8 real-world applications. This program emphasizes practical skills with over 20 Frontier and Open models, catering to developers and AI enthusiasts aiming to enhance their expertise in AI technologies.
