List of Supported Operators for NPU Acceleration

FuriosaAI NPU and SDK can accelerate the following operators, as supported by Tensorflow Lite model and ONNX.

The names of the operators use ONNX as a reference.

Note

If NPU acceleration is not supported, it will run on the CPU. For some operators, even if NPU acceleration is supported, if certain conditions are not met, they may be split into several operators or may run on the CPU. Some examples of this would be when the weight of the model is larger than NPU memory, or if the NPU memory is not sufficient to process a certain computation.

Operators Accelerated on NPU

Name of operator	Additional details
Add
AveragePool
BatchNormalization	Acceleration supported, only if after Conv
Clip
Concat	Acceleration supported, only for height axis
Conv	Acceleration supported, only for group <= 128 and dilation <= 12
ConvTranspose
DepthToSpace
Exp
Expand
Flatten
Gemm
LeakyRelu
LpNormalization	Acceleration supported, only for p = 2 and batch <= 2
MatMul
MaxPool
Mean
Mul
Pad
ReduceL2
ReduceSum
Relu
Reshape
Pow
SpaceToDepth	Acceleration supported, only for mode=”CRD” and Furiosa SDK version 0.6.0 or higher
Sigmoid
Slice	Acceleration supported, only for height axis
Softmax	Acceleration supported, only for batch <= 2
Softplus
Sub
Split
Sqrt
Transpose
Unsqueeze