On April 7, the global authoritative AI benchmark MMLPerf released the latest list, in the focus on low power consumption, high energy efficiency IoT field Tiny v0.7 list, based on the flathead brother Gentetsu RISC-V C906 processor software and hardware joint optimization scheme, achieved the first place in all 4 indicators. This means that in the AIoT space, risc-V architectures can implement customized AI functions at a very low computational cost.
(Picture: Screenshot of MLPerf website)
MLPerf Tiny is currently the authoritative AI list for software and hardware performance and optimization capability testing in the global IoT field, including four typical AI tasks such as visual wake-up, image classification, voice wake-up and anomaly monitoring. This year, the PARTICIPATING CPUs cover ARM, RISC-V and self-developed architectures.
In the Tiny v0.7 list, without the use of accelerators, Ali based on the software and hardware performance optimization results submitted by the Flathead Gentee RISC-V C906 processor, while meeting the accuracy requirements, it refreshed the record of all 4 benchmark indicators, creating the best results of the RISC-V architecture in the AI benchmark.
(Picture: Comparison of performance data of the authoritative AI list MPERf Tiny V0.7)
Relying on Alibaba's software and hardware integration innovation in algorithms, compilation and hardware, the Flathead Brother Gente RISC-V processor has achieved performance optimization in the field of AIoT.
At the algorithm level, Alibaba Cloud Aurora Heterogeneous Acceleration Platform uses SinianML, an architecture-aware model optimization tool, to greatly improve the computational efficiency of AI inference through optimizations such as compression, network structure search, distillation, and auto scaling. At the same time, combined with the domain knowledge of Dharma Academy in speech and visual AI algorithms, it is accelerated through algorithm optimization on specific tasks.
At the compilation and hardware level, based on the industry's earliest mass-produced vector-extended RISC-V instruction set of Gentetsu C906 processor, Pingtou Brother further optimized the neural network model deployment toolset HHB and acceleration library CSI-NN2, and better adapted the AI operator with the hardware, so that the Gentetsu CPU achieved performance upgrades. Both HHB and CSI-NN2 are open source.
It is understood that the Gentetsu RISC-V processor covers various scenarios such as low power consumption, high energy efficiency, and high performance, and supports multiple operating systems, which are widely used in smart home appliances, vehicles, industrial control, edge computing and other fields. Previously, Pingtou Brother had open sourced 4 mass-produced RISC-V processor IP, as well as full-stack software and tools based on Gentetsu.
"The RISC-V architecture is more flexible and more able to meet the customization needs of the AIoT era. The 4th place in the MLPerf Tiny list demonstrates the outstanding potential of RISC-V in terms of performance and energy efficiency. Pingtou Brother will continue to promote the integration and innovation of RISC-V software and hardware technology, continue to contribute to the RISC-V ecosystem, and make RISC-V grow into a new choice for more scenarios and industries. Meng Jianyi, vice president of Pingtou Brother, said.