Skip to content

NPU Performance Introduction*

  • NPU 200M, SNPU 120M, tested with npu_compiler version 1.0.14

1. ASR Model*

  • (FC320 LSTM400 LSTM400 LSTM400 LSTM400 FC211)

    Time Memory Size
    NPU 2.4ms 2.7M
    SNPU 5.8ms 3.4M

2. Lenet5*

  • (1x28x28 -> Conv2D 5x5x1x32 kernels -> Relu -> MaxPool -> Conv2D 5x5x32x64 kernels -> Relu -> MaxPool -> FC 3136x1024 weights -> FC 1024x10 weights)

    Time Memory Size
    NPU 7.4ms 4.6M
    SNPU 13.6ms 4.8M

3. AlexNet*

  • (3x112x112 -> Conv2D 11x11x1x64 kernels -> Relu -> MaxPool -> Conv2D 5x5x64x192 kernels -> Relu -> MaxPool -> Conv2D 3x3x192x384 kernels -> Relu -> Conv2D 3x3x384x384 kernels -> Relu -> Conv2D 3x3x384x256 kernels -> Relu -> MaxPool)

    Time Memory Size
    NPU 221ms 35.1M
    SNPU 359ms 36.1M

4. MobileNet V2 (Model Structure as follows)*

npu_mobilenet_v2

Time Memory Size
NPU 502ms 36.7M
SNPU 1019ms 38.1M