RK3399pro + K7/A7 FPGA + AI Domestic Artificial Intelligence Image Processing Platform
rk3399pro Introduction The rk3399pro is Rockchip's new ARM chip with an NPU. Before its release, the NPU's computing power was stated as 2.4 TOPS, but actual testing after its discovery reached 3.0 TOPS. With such powerful computing capabilities, the Jetson Nano's computing power is 0.47 TFLOPS. Since the units are different, the only way to compare their computing power is through actual network computations.

rk3399pro Computing Power Test
Test Conditions
Hardware: TB-RK3399Pro 3GB RAM + 16GB Flash Storage
Toolkit Version: toolkit 1.0.0
Test Results
Test Conclusion
YOLOv3's performance is very strange; however, in the forum, toolkit 0.9 could achieve 30fps, but this is only 10fps. It's possible that some modifications were made to the network in the official test data.
Overall, the rk3399pro's computing power is very strong, much better than I expected. It can be used directly without significant network modifications. However, this is also its drawback: it might not be able to convert some custom network layers, and we'll need to wait for updated versions.
