Hello everyone, we are a team from Seeed Studio. Recently, we have tested two demos of YoloV8s on Pi5 and CM4, using Hailo 8L for acceleration. We have measured parameters such as frame rate, power consumption, and temperature, and created benchmark charts for easy reference. As a comparison, we also tested the NVIDIA Jetson Xavier NX 16GB using the same demo and input.
Here are a few interesting results:
1- The acceleration effect of Hailo is very noticeable.
2- The frame rate of Pi5 under PCIe gen3 is twice as high as under PCIe gen2.
3- The performance of CM4 after acceleration is unexpectedly good.
4- The power consumption performance of Pi5 and CM4 is excellent when accelerated by Hailo8L
Testing conditions
Input: 240 FPS video. 640*640 Pixels
Same model (YOLOv8s), quantized to int8,
![Image]()
For more details, you can check our wiki link.
https://wiki.seeedstudio.com/benchmark_ ... pi_ai_kit/
As part of our plans, we also tested Google Coral. Unfortunately, after applying acceleration, the Pi5 running the same demo with Coral only achieved 1-2 frames per second. We suspect that there might be a configuration issue, so we haven't uploaded the results for Coral at the moment. Once we have debugged the issue, we will provide the complete results for Coral. We will also test the performance differences and potential loss when using dual PCIe M.2 hats, with SSD and Hailo 8L, under both PCIe 2.0 hub and PCIe 3.0 hub (Yes, we will be releasing a dual M.2 hat for Pi5 with PCIe 3.0 hub soon.).
If you find any issues with the testing methodology or if there are specific projects related to Pi+AI that you would like to test, please feel free to discuss and leave a message. Your feedback is welcome.![:D]( "Very Happy")
Here are a few interesting results:
1- The acceleration effect of Hailo is very noticeable.
2- The frame rate of Pi5 under PCIe gen3 is twice as high as under PCIe gen2.
3- The performance of CM4 after acceleration is unexpectedly good.
4- The power consumption performance of Pi5 and CM4 is excellent when accelerated by Hailo8L
Testing conditions
Input: 240 FPS video. 640*640 Pixels
Same model (YOLOv8s), quantized to int8,
For more details, you can check our wiki link.
https://wiki.seeedstudio.com/benchmark_ ... pi_ai_kit/
As part of our plans, we also tested Google Coral. Unfortunately, after applying acceleration, the Pi5 running the same demo with Coral only achieved 1-2 frames per second. We suspect that there might be a configuration issue, so we haven't uploaded the results for Coral at the moment. Once we have debugged the issue, we will provide the complete results for Coral. We will also test the performance differences and potential loss when using dual PCIe M.2 hats, with SSD and Hailo 8L, under both PCIe 2.0 hub and PCIe 3.0 hub (Yes, we will be releasing a dual M.2 hat for Pi5 with PCIe 3.0 hub soon.).
If you find any issues with the testing methodology or if there are specific projects related to Pi+AI that you would like to test, please feel free to discuss and leave a message. Your feedback is welcome.
Statistics: Posted by Ainsfei — Thu Jul 18, 2024 10:44 am