破解石油石化数智化转型难题，海光信息给出中国“芯”答案

Core Insights - The energy sector is accelerating its digital transformation from "point breakthroughs" to "system reconstruction" and from "technology application" to "ecosystem co-construction" under the guidance of multiple strategies like "Artificial Intelligence+" [1] Group 1: Industry Trends - The oil and petrochemical industry, as a pillar of the national economy, is becoming a core area for digital transformation due to its rich AI application scenarios [2] - A heterogeneous computing system composed of various chips has become mainstream, but insufficient interconnect efficiency is hindering deeper intelligence development in the industry [2] - The industry consensus is that addressing the bottleneck of efficient collaboration between heterogeneous chips is a key challenge [2] Group 2: Company Solutions - Haiguang Information showcased its "CPU+DCU" heterogeneous computing architecture solution, which significantly enhances processing efficiency by providing bandwidth far exceeding PCIe and lower latency [2] - The "CPU+DCU" solution, powered by the DeepAI intelligent engine, enables full-chain optimization from basic operators to upper-layer applications, lowering the AI application threshold with over 100 scenario modules and 300 fine-tuned models [2] - Haiguang's solutions have accumulated rich practical experience in the petrochemical industry [2] Group 3: Case Studies - Sinopec has adopted Haiguang's "CPU+DCU" liquid cooling computing cluster to support over 200 AI scenarios across nine categories, significantly improving exploration success rates, production efficiency, and reducing safety incident rates [3] Group 4: Security Challenges - The collaboration between CPU and GPU in the oil and petrochemical industry faces security challenges, including tightly coupled security and business functions and a lack of unified security standards [4] - Haiguang's CSV3.0 technology establishes a hardware-level secure data transmission channel between CPU and DCU, effectively defending against side-channel attacks and ensuring data is "available but not visible" throughout processing [4] - The company aims to solve issues like scattered application scenarios and data silos by building a comprehensive technology ecosystem with over 6,000 partners [4]