Huawei Technologies on Friday introduced open source software that it claims can significantly improve the utilization of artificial intelligence chips, marking the Chinese company’s latest effort to deliver world-class AI training capabilities even without access to the latest Nvidia processors.
Flex:ai, which pools and tunes processors to improve performance, comes as part of Huawei’s efforts to overcome China’s shortcomings in individual chipsets and develop a self-sufficient computing ecosystem.
The software is built on an open-source platform called Kubernetes, which is an orchestration system for graphics processing units (GPUs), neural processing units (NPUs), and other accelerators from various chip manufacturers. According to Huawei, a single card can be divided into multiple virtualized computing units to run multiple AI workloads in parallel.
The company says the tool can improve processor utilization by an average of 30%. A smart scheduler called Hi Scheduler handles cluster-wide GPU and NPU allocation. The system can also pool idle processors from different nodes and redistribute that power to AI tasks.
Currently the backbone of AI infrastructure, Kubernetes is an open source system that orchestrates large fleets of containerized applications. Containerization is a technique that makes it easy to deploy and scale AI models across different servers and GPU clusters.
“Small tasks rarely use up the capacity of one card, but large tasks cannot be handled by just one card, and parallel tasks make computing management difficult,” Zhou Yuefeng, Huawei’s vice president and head of data storage product line, said Friday at a launch event at Shanghai’s Lianqiuhu campus.
Flex:ai is designed to improve the utilization of computing resources and is part of Huawei’s efforts to accelerate the “democratization of AI” by unlocking the potential of AI infrastructure and open sourcing, he said.
