[Detector Support]: Is it possible to run CoralTPU and Nvidia TensorRT at same time? If so, how? Example configs crash when combined #11244
Unanswered
douginoz
asked this question in
Detector Support
Replies: 1 comment
-
No, this is not supported |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Describe the problem you are having
I have 2 pci-e Coral TPUs that are configured and work fine in Frigate. I now want to try adding TensorRT detection in addition to the existing 2 Corals. Doing this kills frigate. If I remove the CoralTPUs from the config, the TensorRT stuff works as expected. If I remove TensorRT stuff, the Corals work as expected. But trying to have them all configured crashes things.
Frigate 0.13.2-6476F8A
go2rtc 1.8.4
ubuntu 22.04
GPU NVidia GeForce GTX 1660 Super
Nvidia driver 0.13.2-6476F8A
CUDA Version 12.2
This works:
And This works:
But this doesn't:
I have also tried that last one with the entire 'model:" section removed but makes no difference.
Attached are 3 logs:
A - With dual coralTPUs configured (using first config snippet shown above)
B - With TensorRT configured only (using second config snippet shown above)
C - With dual CoralTPUs and TensorRT configured (using thirt config snippet shown above)
I'm guessing that it's related to the 'model:" section, which appears to be required for tensorRT, but by including it in the config it confuses frigate's attempt to initialize the CoralTPU detection. So this leads to the question:
Is it possible to have tensorRT detection AND CoralTPU detection at the same time, and if so, what's the config trick to make them both work?
A - Dual Corals Only.log
B - TensorRT Only - with model.log
C - Dual Corals and TensorRT.log
Version
0.13.2-6476F8A
Frigate config file
docker-compose file or Docker CLI command
Relevant log output
Operating system
Debian
Install method
Docker Compose
Coral version
M.2
Any other information that may be helpful
I've only tried adding in TensorRT tonight. In doing so I changed the docker-compose to use
image: ghcr.io/blakeblackshear/frigate:stable-tensorrt
instead of
image: ghcr.io/blakeblackshear/frigate:stable
It spent a few minutes generating the new files needed. I have checked within the docker instance that the tensorRT model is available at /config/model_cache/tensorrt/yolov7-320.trt
Beta Was this translation helpful? Give feedback.
All reactions