GSI Technology announced that its Gemini‑II compute‑in‑memory processor achieved a 3‑second time‑to‑first‑token (TTFT) when running the 12‑billion‑parameter Gemma‑3 vision‑language model at a power draw of 30 W. The benchmark demonstrates the fastest TTFT among publicly reported embedded edge processors, positioning GSI as a strong contender in the battery‑constrained edge‑AI market.
The Gemini‑II’s 3‑second TTFT matches NVIDIA’s Jetson Thor benchmark (≈3 s) but does so at a fraction of the power—30 W versus >100 W—while outperforming Qualcomm’s Snapdragon X Elite, which reports ≈12 s TTFT at the same 30 W. This power advantage is critical for applications such as drones, smart‑city cameras, and defense sensors, where thermal limits and battery life dictate deployment feasibility.
The performance gain stems from GSI’s compute‑in‑memory (CIM) architecture, which integrates processing directly into memory arrays, dramatically reducing data movement and associated energy costs. The CIM design enables the Gemini‑II to deliver low latency without the high power consumption typical of conventional CPU‑GPU pipelines, a key differentiator in the edge‑AI space.
GSI’s financial context underscores the strategic importance of this milestone. In fiscal Q2 2026, the company reported revenue of $6.4 million—up from $4.6 million a year earlier—while posting a net loss of $0.11 per share. Management guided Q3 revenue to $6.0 million–$6.8 million, reflecting confidence in continued demand for high‑margin, mission‑critical contracts. The company is actively collaborating with G2 Tech on a government‑backed autonomous perimeter‑security system, illustrating early commercial traction in defense and smart‑city segments.
CEO Lee‑Lean Shu emphasized that “edge deployments require fast response under tight power and thermal limits,” and that the benchmark “highlights what compute‑in‑memory can enable for physical AI.” GSI continues to refine Gemini‑II’s responsiveness and seeks to translate the technical achievement into commercial success, targeting drones, defense, and smart‑city markets while maintaining a focus on power‑efficient, low‑latency solutions.
The content on EveryTicker is for informational purposes only and should not be construed as financial or investment advice. We are not financial advisors. Consult with a qualified professional before making any investment decisions. Any actions you take based on information from this site are solely at your own risk.