Use GPU+2xDLA in deepstream app config

• Hardware Platform Jetson NX
• DeepStream5.0
• JetPack Version 4.4
• TensorRT 7.1
• Issue Type: questions
How can i run deepstream app with GPU+2xDLA in YoloV3 Tiny model to good fps like Jetson Benchmarks

Hi,

Please noted that one model could only deploy on a process per time.
The link’s performance is measured by overall throughput, which means three identical models run on GPU and DLAs.

If you want to do this with Deepstream, you will need to duplicate the pipeline for a different processor.

Thanks.

1 Like

deepstream-app -c gpuconfig -c dla0config -c dla1config

i found this comand at https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Performance.html#running-applications-using-dla, i don’t understand to use it

Hi,

To run GPU+2xDLA in the same process, you will need three configure files for each processor.
Three configure files but different in the processor setting.

GPU

[property]
...
enable-dla=0

DLA 0

[property]
...
enable-dla=1
use-dla-core=0

DLA1

[property]
...
enable-dla=1
use-dla-core=1

And launch the deepstream-app with the above configure files.

$ deepstream-app -c [path/to/gpu/configure/file] -c [path/to/dla0/configure/file] -c [path/to/dla1/configure/file]

Thanks.

1 Like

thank you