Huawei’s new algorithm could cut China’s reliance on foreign memory chips
Summary
Huawei unveiled a new algorithm called Unified Cache Manager at a forum in Shanghai. The company says the software improves inference efficiency by allocating data across different memory types based on latency needs. Huawei claims the algorithm cut inference latency by up to 90% and boosted throughput by as much as 22 times in tests. The firm plans to open-source the technology in September, first through its developer community and then to the broader industry.
Classifications
industries
No industries detected
applications
ERP & Process Management
AskAI Classifications
Labels
Cloud Computing
Developer Tools
Enterprise Software
