AT the LiveVideoStack Audio and Video Technology Conference in December, Huawei Cloud released a solution in the field of visual computing: the Shengteng AI computing solution. According to reports, the Shengteng AI computing solution is based on Shengteng’s full-stack self-research capabilities, builds an open development and service framework, provides rapid integration, shares standardized components, and achieves efficient tuning. Industries such as pan-financial certification and smart healthcare provide full-stack empowerment. Full coverage of AI visual computing scenes: Computing power • Equipped with the Shengteng 310 chip, real-time inference performance is improved by 2x times (Inception-V4: BS = 1) • Cloud Instance Comprehensive Scenario improves cost performance by 2x times Cloud-edge-end integration AI architecture • Distributed AI cluster: AI container computing cluster, performance increased by 30% • Edge Cloud Collaboration AI • Edge Collaborative AI Full stack self-developed, support the whole process open • Ascend Serving-Standardization of Shengteng service components, supporting interfacing with OBS, SFS-Turbo, supporting containerized deployment, and edge deployment. • Ascend Graph-Ascent components, business process communities and markets, allowing developers to quickly build applications based on Ascent’s computing power and accelerate business model deployment. • MindSpore: Support unified, independent and collaborative training and inference framework for end, edge and cloud, and optimize inference performance through optimization of compilation parameters. |