UINT8 model could not run on the vim3 NPU!


  1. Can PTH or PT models with pytorch quantified by pytorch be directly run on rk3399 arm and rk3399pro NPU?
  2. If I use other formats of quantitative models, such as onnx, how can I run on rk3399 and rk3399rpo npus?
  3. what formats of quantitative models does rk3399pro NPU support? Or does it mean that the model must be converted to the rockchip framework before further quantification and finally deployed on the board?


@lcl2020 The main chip of our Edge series is rk3399, rk3399 does not have NPU. We do not have rk3399pro related boards, so there is no support for rk3399pro.

@lcl2020 I didn’t understand what you meant

How could I know if my vim3 board especially a311d NPU tuned on the high power mode or low power mode? I mean the NPU will show the best performance with cnn model?

Any info could show that ?

@lcl2020 To ensure that the CPU will not reduce the frequency to maintain the performance of the NPU

you mean the cpu freq is proportional to the freq of npu, right?

If at low power mode ,what should the BogoMIPS of arm cpu freq be?
what if at high mode? tks

In fact , RK 1808 1109 1126 is all right? Do you have these boads for NPU inference?

@lcl2020 We only provide technical support for our boards. Our Edge series does not come with an NPU

any plan for rk3399pro rk28xx or rk25XX