![]() ![]()
This is really exciting, we can wait to get our hands on this SoC and put it into various applications. We also noticed, there is QSGMII words in the reference design though it’s not implemented, which means in the future we may connect 5Gbps ethernet phy to RK3566. It has very rich interfaces and many high speed peripherals. Support one-click model conversion, support Caffe/TensorFlow/TF-Lite/ONNX/PyTorch/Keras/Darknet mainstream framework modelsĪlthought RK3566 is target as a middle range SoC, but we think the performance is good enough for many applications, from multimedia human-machine interaction, to computer vision to robotics.NPU internal blocks include multiply-add operations, activation, LUT and precision conversion units, and support custom layer construction.The NPU core supports ordinary convolution, depth separable convolution, deconvolution, hole convolution, fully connected layer and pooling layer.Support lossless compression of INT8, INT16, FP16 neural network parameters.NPU hardware natively supports technologies such as pre-processing merging, channel quantization, and zero skipping.Embedded neural network hardware accelerator, support INT8, INT16, FP16 efficient operation.Hightlight 4: Integrated efficient RKNN AI processing unit Support dynamic bit rate, frame rate, resolution adjustment. Rockchip driver windows 10 cpu 322a 1080p#
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |