According to @godofprompt, GSI Technology has introduced the Gemini-I chip, which matches the performance of Nvidia’s A6000 GPU while using 98% less energy, as validated by a Cornell University study (source: Cornell University, Twitter). The chip leverages compute-in-memory architecture, integrating processing directly inside memory arrays to eliminate the energy cost of data transfer. In rigorous benchmarks on real-world AI workloads, including retrieval-augmented generation tasks for chatbots, Gemini-I ran five times faster than standard CPUs and consumed only 1 to 2% of the energy required by GPUs (source: Cornell University study). This breakthrough could dramatically cut data center power consumption, enable edge AI in power-constrained environments, and reduce AI’s overall climate impact. The $100 billion AI inference market faces disruption as GSI’s approach delivers GPU-class performance with drastically reduced energy costs (source: Twitter, Nature). Other industry leaders like MediaTek are adopting similar compute-in-memory techniques, highlighting a shift toward energy-efficient AI hardware.