Performance Analysis of RK3568, Rockchip New Generation AIoT Processor
Rockchip's new generation AIoT processor RK3568, a professional universal SoC based on advanced 22nm processing technology, could be widely used in Industrial Internet, HMI, NVR storage, vehicle central control, industrial gateway, etc.
All in one--CPU, GPU, NPU and VPU
quad-core 64-bit Cortex-A55
up to 2.0GHz
advanced 22nm processing technology
Supports OpenGLES 1.1/2.0/3.2, OpenCL2.0, Vulkan1.1
Embedded high-performance 2D acceleration hardware
Integrated powerful AI accelerator RKNN NPU
Up to 1 TOPS
One button convert to TensorFlow//Caffe, etc.
Decode H.264 4K@30fps, H.265 4K@60fps
Encode H.264/H.265 1080P@60fps
8M ISP, supports HDR
Function comparison between RK3568 VS RK3399
RK3399, also by Rockchip, is a low-power, high-performance application processor based on Big.Little architecture, which combines dual-core Cortex-A72 with quad-core Cortex-A53 by independent NEON co-processor. Target application in computers, personal Internet mobile devices, VR, advertising machines, etc.
|CPU||Quad-core Cortex-A55 2.0GHz||Dual core Cortex-A72 + Quad-core Cortex-A53|
|NPU||Up to 1Tops||None. But can connect to 1808 computing stick|
|Media||H264,H265,VP8,VP9 video decode, up to 4K@60fps H265 JPEG decode.||H264, H265, VP8, VP9 video decode, up to 4K VP9 and 4K 10bits H265/H264|
|Display||Support, HDMI, MIPI/eDP, LVDS, three displays simultaneously output||Support HDMI, DP, eDP, LVDS, MIPI, etc two of above displays simultaneously output|
|Network||Ethernet*2, 4G/5G||Ethernet*1, 4G/5G|
|Interface||USB2.0*2, USB3.0, Type-C, PCIe3.0, PCIe2.1, SATA3.0, QSGMII, CAN, etc.||USB2.0*4, USB3.0, Type-C, M.2 PCIe, etc.|
From above comparison, you can see both RK3568 and RK3399 are quite excellent universal SoC. However, RK3568 has a high-performance 2D acceleration hardware, which supports JPEG decode, 2 Ethernet ports, three displays simultaneously output, multi PCIE\SATA interfaces. Coupled with the NPU, up to 1Tops, RK3568 has a impressed performance in image processing, storage, communication, and multi-function peripherals.
How to performance of RK3568?
About FET3568-C SoM
It carries on RK3568 with rich functional interfaces, powerful multimedia performance, low power consumption, and stable quality. It can be widely used in Power, Medical, IoT, Industrial, Transportation, and AI.
1. Display Test
Same display on three screens
Different display on three screens
Simultaneous play on MIPI/LVDS/HDMI
Video hardware decode
Play 1024*600/1280*800/3840*2160 three resolution videos at the same time.
Type commands for H.264 video hard-decode play
[root@ok3568:/]# gst-launch-1.0 filesrc location=/home/forlinx/video/1-B.mp4 ! qtdemux ! h264parse ! mppvideodec ! waylandsink "render-rectangle=<0,0,1280,800>" &
[root@ok3568:/]# gst-launch-1.0 filesrc location=/home/forlinx/video/2-3001.mp4 ! qtdemux ! h264parse ! mppvideodec ! waylandsink "render-rectangle=<1280,0,3840,2160>" &
[root@ok3568:/]# gst-launch-1.0 filesrc location=/home/forlinx/video/3-Q.mp4 ! qtdemux ! h264parse ! mppvideodec ! waylandsink "render-rectangle=<5120,0,1024,600>" &
As shown above, when play under 1024*600/1280*800/3840*2160, the CPU usage is only about 30%, 70% left for other operations.
2. Hardware codec
Rockchip official mpp hardware decoding components, H.264\H.265\VP8\JPEG, etc.
Use the command to view the inspect as follows:
[root@ok3568:/]#gst-inspect-1.0| grep rockchipmpp
For more details, you can refer to Forlinx User Manual.
RK3568 supports 4K60 frame video decoding, and can equivalently decode multiple video sources at the same time. The decoding efficiency is significantly improved. Besides, RK3568 has a dedicated hardware JPEG decoding processor, 240MPixels per second, and supports concurrent processing of multiple small images. Compared with most similar products by software decoding, RK3568 is more efficient.
Built-in Tensorflow AI framework and Open CV3.43 basic library, convenient for users to use directly.
Test Tensorflow on FET3568-C System on module under Linux, like following:
[root@ok3568:/]# cd /usr/local/tensorflow
[root@ok3568:/usr/local/tensorflow]# chmod +x label_image_mobilenet
[root@ok3568:/usr/local/tensorflow]# ./label_image_mobilenet -m mobilenet_v1_1.0_224.tflite -l labels.txt -i cat.bmp -a 0 -c 100
[root@ok3568:/usr/local/tensorflow]# chmod +x label_image_mobilenet_ssd
[root@ok3568:/usr/local/tensorflow]# ./label_image_mobilenet_ssd -m mobilenet_ssd.tflite -i test.bmp -a 0 -c 100
Run the face test DEMO by Forlinx Embedded
[root@ok3568:/]# fltest_opencv_rknn_ssd /userdata/model/ssd_inception_v2.rknn /userdata/model/road.bmp
With up to 1Tops NPU and high-performance AI capacity, it can develop more intelligent applications at the edge and reconstruct edge value.
4. Multimedia Performance Test
Gstreamer audio and video playback platform, UVC Camera and MIPI OV13850 camera, support video recording and playback in various formats.
Camera takes picture
[root@ok3568:/]# gst-launch-1.0 v4l2src device=/dev/video0 num-buffers=1 !video/x-raw,format=NV12,width=640,height=480 ! mppjpegenc !filesink location=pic.jpg
H.264 video recording
[root@ok3568:/]# gst-launch-1.0 v4l2src device=/dev/video0 num-buffers=100 !video/x-raw,format=NV12, width=640,height=480 ! tee name=t ! queue !mpph264enc !queue !h264parse !qtmux !filesink location=13850_h264.mp4 t. ! queue !waylandsink
RK3568 has built-in 8MISP image signal processor, which support dual camera and HDR.
The video input interface can be connected to an external camera or expand the input capability of multiple cameras. Thus can meet users’ different requirements in multimedia.
RK3568 has GPIO interfaces, PCIe bus, CAN bus, RTC, onboard MIC, onboard MIPI CSI, headphone jack, HDMI, MIPI, LVDS, 4G/5G, 2 Ethernet ports, etc. Here take CAN bus as an example.
Test of CAN bus interface
2-ch CAN bus interface, connection method: connect H terminal of CAN to the H terminal of other CAN devices, and L terminal of CAN to the L terminal of other CAN devices. Short-circuit CAN0 and CAN1, and type the following command on development board:
CANFD supports up to 64 data bytes per data frame.
Set CAN0/CAN1, arbitration bit to 1M and data bit to 5M.
[root@ok3568:/]# ifconfig can0 down
[root@ok3568:/]# ifconfig can1 down
[root@ok3568:/]# ip link set can0 up type can bitrate 1000000 sample-point 0.75 dbitrate 5000000 dsample-point 0.8 fd on
[root@ok3568:/]# ip link set can1 up type can bitrate 1000000 sample-point 0.75 dbitrate 5000000 dsample-point 0.8 fd on
[root@ok3568:/]# ifconfig can0 up
[root@ok3568:/]# ifconfig can1 up
can0 The device is the server (the server first executes the following commands)
[root@ok3568:/]# candump can0&
can1 The device acts as a client (the client sends data)
[root@ok3568:/]# cansend can1 123#1aabbccddaabbccd
can0 123  1A AB BC CD DA AB BC CD
RK3568 has a more flexible IOMUX and pin multi-function combination application, to meet more diverse functional combination requirements. Users can select according to their product design.