Blue Iris 6.1.3.6 - TensorRT-RTX (GPU) reverts to DirectML after reboot

I

indi1984

n3wb
Mar 29, 2022
13
11
Cleveland, Ohio
Hello all,

I am giving Blue Iris built in AI another shot after learning some Nvidia control panel tricks (specifically to force no fallback to system memory for BlueIris.exe service which forces the use of my RTX3050 vram instead of system memory). One issue that I can not figure out is that TensorRT reverts to DirectML after every reboot. I have clean installed the latest Nvidia studio driver for my RTX3050, I have reinstalled Cuda toolkit 12.4 and check that the PATH was correct in env variables, I have reinstalled Blue Iris from scratch, I have added a service dependency to the BlueIris system service so that it waits for the NVidia service to fully start before starting itself. Nothing I have tried has remedied the issue. It works when I manually set to TensorRT after boot. I run Nvidia-smi, and I get the blueiris.exe service at "Type C" and according to system monitor my RTX3050 is being used as well as its vram shows useage. So it doesn't appear that blue iris is falling back to DirectML at all..... again until after I reboot the machine.
 
indi1984 said:
Spent about 2 hours reading everything I could. No luck getting TensorRT-RTX to stick after reboot. I just let it revert to DirectML and am leaving it at that for now. It works and gives me good inference times. Thank you.
Click to expand...
You should probably send in a bug report to Ken, the BlueIris developer. I have the same thing happening, but without a "real" GPU (slot/card) installed: my system uses the built-in Intel 630 GPU and AI analysis runs much faster if using the "OpenVINO (GPU)" driver, but it also has a strong tendency to revert to using the one of the DirectML drivers after a reboot (which work, but quite a bit slower).
 
  • Like
Reactions: VorlonFrog
jrbeddow said:
You should probably send in a bug report to Ken, the BlueIris developer. I have the same thing happening, but without a "real" GPU (slot/card) installed: my system uses the built-in Intel 630 GPU and AI analysis runs much faster if using the "OpenVINO (GPU)" driver, but it also has a strong tendency to revert to using the one of the DirectML drivers after a reboot (which work, but quite a bit slower).
Click to expand...
Sent in a bug report to Ken this morning. Going back and forth on testing currently. Thanks!
 
  • Like
Reactions: VorlonFrog and jrbeddow
You must log in or register to reply here.
Top Bottom