Sugon(SH:603019) Huan Qiu Wang·2025-11-06 10:17Core Insights - The launch of the scaleX640 super node by Zhongke Shuguang represents a significant advancement in computing infrastructure, addressing the bottlenecks in large-scale deployment and commercial usability of intelligent computing systems [1][2] - The scaleX640 product is designed to support 640 accelerator cards in a single cabinet, achieving a 20-fold increase in computing density compared to similar products, and a 30% to 40% performance improvement in training and inference scenarios for trillion-parameter models [2] Group 1 - The scaleX640 features a "one-to-two" high-density architecture, enabling ultra-high-speed interconnection within a single cabinet [1] - The product integrates multiple technologies, including ultra-high-speed orthogonal architecture, high-density blades, immersion phase change cooling, and high-voltage direct current power supply, achieving global optimization of hardware and software [1][2] - The product has completed over 30 days of stable operation testing, ensuring reliability for future deployments of super-large-scale clusters with tens of thousands of cards [2] Group 2 - Zhongke Shuguang has collaborated with over 20 enterprises in the industry to launch the "AI Computing Open Architecture," addressing challenges such as insufficient high-end chip computing power and lack of unified standards [2] - The scaleX640 is a practical outcome of this architecture, supporting multi-brand accelerator cards at the hardware level and being compatible with mainstream computing frameworks at the software level [2] - The release of scaleX640 is significant for both breaking through computing bottlenecks and providing efficient, user-friendly intelligent computing infrastructure for continuous innovation in large models and AI applications across various industries [2]