The first domestic computational power center-grade inference acceleration card is officially released in Shenzhen. Speed increases by 40% and energy consumption decreases by 25%.

On November 11th, during the celebration of the relocation of the Jiangyuan Technology headquarters to the Shenzhen Luohu Chengjian Cloud Start Building, the first domestically produced national-level inference accelerator card - the Jiangyuan D20 AI accelerator card was released simultaneously. It is reported that this fully domestic product adopts a dual-chip architecture, integrating two domestically produced artificial intelligence chips developed independently by Jiangyuan, doubling the card's computing power performance and memory capacity. The authoritative institution, the Electronic Four Institute's Saisi Laboratory, has completed comprehensive testing of the D20 AI accelerator card, and its specifications such as computing power, storage, interconnection, and power consumption have all passed authoritative certification. Its application deployment capabilities for traditional AI models and large models such as Qwen3-235B have also been officially recognized. Test data shows that when running mainstream large models, the response speed of this product increases by 40% and energy consumption decreases by 25%. It can be widely adapted to various industries such as smart manufacturing and financial technology, and has already been successfully delivered for commercial use.