Huawei Technologies has announced its aim to significantly increase the utilization of artificial intelligence chips with its newly introduced open-source software called Flex:ai. The company emphasized that this move is intended to compensate for China's restrictions on access to the most advanced processors.
The editorial office reports that Flex:ai has been introduced as an orchestration system that combines GPUs, NPUs, and other accelerators from various manufacturers, ensuring their full load operation. The software can divide a single physical card into several virtual computing units, enabling the parallel execution of multiple artificial intelligence tasks.
Huawei states that the software is capable of increasing the actual utilization rate of chips by an average of 30 percent. The management mechanism included in the system, called Hi Scheduler, optimizes resource allocation across clusters and can also gather idle processors and redirect them to other workloads.
The company announced that the Flex:ai project will be openly shared through the Huawei ModelEngine community. Specialists from Shanghai Jiao Tong, Xi'an Jiaotong, and Xiamen universities also participated in the research.
Experts note that the software's functionality is reminiscent of the technology from Run:AI, a company acquired by Nvidia in 2024. This indicates that Chinese companies' strategy focused on resource optimization aligns with global trends.