Multi-Drvice/Pipeline Host Ram Usage

bherbruck

I think depthai multiplies ram usage for pipelines in a Multi-Drvice setting, for each device my ram usage goes up by about the same size as my pipeline's yolo model after the pipeline is sent to the device - even with no xlink out nodes in the pipeline.

Is there a way to clean up pipeline resources after they are sent to the device?

jakaskerl

bherbruck
cc @MatevzMorato

bherbruck

Quick clarification, even when not flashing a real pipeline and just running device.startPipeline() (blank pipeline, no output queues) for each device in its own thread, I'm up to ~500MB of ram for 8 devices just idling.

jakaskerl

bherbruck
If you do pipeline=None after starting the pipeline, does the memory get freed?

Thanks,
Jaka

bherbruck

jakaskerl on my latest test I didn't even have a pipeline, I just did something like:

with Device(device_info) as device:
    device.startPipeline()
    while True:
        time.sleep(1)

jakaskerl

bherbruck
I get about 30MB per device.. Where are you getting 500MB of usage? How do you manage the context (threading/cm/..)?

Thanks,
Jaka