Can Specific NPU Cores Be Selected for Inference on Genio 1200?

On the Genio 1200 platform, the NPU contains two cores. Is it possible to select which NPU core is used for AI inference, or configure the system to restrict execution to a single core or enable both cores? Please specify the recommended system properties or configurations to control NPU core usage. Any limitations or best practices for core selection would also be appreciated.

MediaTek currently does not support specifying a particular NPU core for AI inference on Genio 1200; however, it is possible to configure single or dual-core operation. Use the following system properties:

  • Enable dual NPU:
    setprop debug.mtk.apu 1
    setprop debug.mtk.apu.multicore.enable 1
  • Disable dual NPU (force single core):
    setprop debug.mtk.apu 1
    setprop debug.mtk.apu.multicore.enable 0
    To restrict to a single MDLA core via Neuron delegate, use:
    setprop debug.neuron.adapter.AdapterCompilerOptions '--num-mdla=1'