GPU memory full with CPAI

[mich2222]

2.png (13.24 KiB) Viewed 4601 times

My hardware:
2x e5-2690v2
32gb ram
gt 1030 2gb

Hi all, I'm using CPAI with blueiris with GPU enabled.
However, I noticed that for the first few minutes everything works normally, after that the 500 error appears in blueiris and the error appears on the CPAI dashboard (Object Detection (YOLOv5 6.2): ​​[Exception] : Traceback (most recent call last): I tried to use both the latest version 2.1.9 and also to downgrade to version 2.0.8 but nothing changed.
After several days spent looking for the cause, I realized that by disabling the GPU and using only the CPU the problem does not occur, furthermore with the GPU everything works until the memory becomes "saturated". It uses full dedicated gpu memory constantly without ever shrinking, and also seems to use traditional ram (shared memory)
Thanks everyone for any advice

[TimG]

Noting "Cam12" in the image, is it possible that you are simply loading the gpu up too much ? I'm using a 4GB GTX1650 here for CPAI and it works well.

[mich2222]

Noting "Cam12" in the image, is it possible that you are simply loading the gpu up too much ? I'm using a 4GB GTX1650 here for CPAI and it works well.

Even disabling all other cameras does not change.

[TimG]

Is that you posting over on the CPAI forum too ? If so I hope they can help, as this is a peer to peer forum, and we are not experts in CPAI.

[MikeBwca]

Getting the same 'AI Error 500'. But only about 5 a day! I can match the error time with an actual alert that has run thru Motion and AI successfully, appears in the Alert List. So, I have no idea what it could be. Memory usqge is normal for Cpu/Gpu.

Previously, I was getting hundreds of AI: error 500's and 3's. Rebooted, and now just getting a few 500's.


These just started happening, with no correlation for an CP or BI update. Seems random.




Had another issue last night... Motion detection just stopped on some, but not all, cameras! Rebooted and deleted the entire DB. Back to normal now.