Why am I seeing the following TensorRT Error in the Processing Server/Alert Processing Server log?
TensorRT Error - Serialization Error in nvinfer1
TensorRT Error: INVALID_STATE: UInknown exception
TensorRT Error: INVALID CONFIG: Deserialize the cuda engine failed
Failed to load engine from cache. In BriefCam::BCDNN
This error occurs when the engine cache is built incorrectly and it crashes the Processing Server/Alert Processing Server services on startup and the process remains stuck in a loop.
To resolve this issue:
Open the Windows services (
services.msc) and stop the BriefCam VSService.Go to the BriefCam server directory and manually delete the
engine_cache_0folder. If there is more than one GPU, there will be more than one engine cache folder. Delete them all.Start the PS/APS by double clicking on
ProcessingServer.exeorAlertProcessingService.exe, which are located atC:\Program Files\BriefCam\BriefCam Server.Let the process build the cache. This can take up to 15 minutes on some GPUs.
Once the process console is idle, close it down and start the BriefCam VSService.
Verify that the logs are not showing a crashing loop anymore.