Computing power surges 10 times

The computing power increased by 10 times! Google secretly develops Frozen chip, hard-core upgrade of Gemini large model. Facing the increasingly severe shortage of AI computing power, Google is secretly developing a new dedicated server chip code-named Frozen v2. The core breakthrough of this chip is to solidify the underlying computing logic of the Gemini large model directly into the hardware, thereby greatly improving the chip's computing efficiency. Hard-core architecture reshapes the computing power ecosystem. According to calculations by the R&D team, the processing performance per unit power consumption of Frozen v2 will be 6 to 10 times that of Google's latest self-developed chips currently in use. This design of etching software algorithms directly into silicon wafers can significantly reduce data handling and logical judgment steps and achieve a qualitative change in energy efficiency. Different from general-purpose chips such as NVIDIA GPU or Google TPU, this dedicated chip is an integrated software and hardware tool specially designed for the Gemini architecture. It can not only greatly shorten AI response time, but also support more complex new application scenarios. It is expected to be deployed as early as 2028. The positioning experiment platform does not replace TPU. Although its performance is strong, Google does not intend to use Frozen chips to completely replace existing TPU chips. The two will complement each other in the future. Due to the solidified algorithm logic, this chip requires the Gemini model to use a specific infrastructure for a long time, so it has a certain technical specificity. Google currently positions it as a technology test platform to accumulate engineering experience for the next generation of more specialized AI chips. Due to the small initial mass production scale, the chip will mainly be targeted at specific internal needs and will not be rolled out on a large scale for the time being. via AI News (author: AI Base)