Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Is it normal to have huge fps loss when I switch yolov5 to use DLA? #601

Open
berkay-karlik opened this issue Dec 10, 2024 · 3 comments
Open

Comments

@berkay-karlik
Copy link

I have followed the exact yolov5 export process described here and I got 9 fps on full gpu.
but when I add/modify the following on config_infer_primary_yoloV5.txt:

enable-dla=1
use-dla-core=0
network-mode=2

My fps drops to 3. In JTOP I observe gpu usage is around %1 and dla0 is definitely getting used. Is it just that dla0 is that underperforming compared to gpu or Did I do something wrong?

Also thanks for the project, great work.

@marcoslucianops
Copy link
Owner

I noticed that, but I'm not using DLA, so I didn't check the issue yet.

@berkay-karlik
Copy link
Author

Okay good to know, just to give more info I have compared ultralytics gpu and dla versions and there is similar amount of difference. After all it might be due to layers that are falling back to gpu. In that case there is nothing to solve. I will be using the gpu too.

@SarperYurttas
Copy link

Screenshot 2025-01-06 at 10 56 41

According to the datasheet of Jetson ORIN NX, the total(GPU+DLA0+DLA2) AI performance is 100(60+20+20) TOPS. Each DLA engine has 20 TOPS individually so the GPU-DLA performance difference should be around 3:1. However in practice, this difference may differ from model to model.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants