GSI Technology’s AI Tech Outperforms GPUs with Massive Energy Savings

SUNNYVALE, Calif., Oct. 20, 2025 (GLOBE NEWSWIRE) — GSI Technology, Inc. (Nasdaq: GSIT) announced a significant advancement in artificial intelligence (AI) processing with the publication of research from Cornell University. The study confirms that GSI’s Associative Processing Unit (APU), designed for compute-in-memory technology, can achieve performance levels comparable to GPUs while dramatically decreasing energy consumption.

Lee-Lean Shu, Chairman and CEO of GSI Technology, stated, “Cornell’s independent validation confirms what we’ve long believed—compute-in-memory has the potential to disrupt the $100 billion AI inference market. The APU delivers GPU-class performance at a fraction of the energy cost.”

The findings were published in the ACM journal and presented at the Micro '25 conference. The research titled “Characterizing and Optimizing Realistic Workloads on a Commercial Compute-in-SRAM Device,” assessed the GSI Gemini-I APU against traditional CPUs and GPUs. It focused on retrieval-augmented generation (RAG) tasks across datasets ranging from 10GB to 200GB.

The Cornell research team found that GSI’s APU consumed over 98% less energy compared to a conventional GPU and reduced processing times by up to 80% compared to CPUs. This efficiency opens new opportunities for industries such as robotics, drones, and IoT devices, where energy use is a critical factor.

Furthermore, GSI’s upcoming Gemini-II APU is expected to provide about 10 times faster throughput and lower latency for memory-intensive workloads, enhancing energy efficiency further. Mr. Shu noted that these advancements will position GSI Technology strongly across various sectors, leveraging the unique speed and programmability of the APU.

The Cornell study additionally introduced an analytical framework that contributes to the optimization of compute-in-memory devices. GSI Technology focuses on translating these innovations into scalable solutions for future applications.