Please provide complete information as applicable to your setup.
• Hardware Platform (Jetson / GPU) Jetson AGX ORIN
• DeepStream Version 6.2
• JetPack Version (valid for Jetson only) JP5.1
As per i know that running same model in 2 DLAs is not possible. But i have found the below comparison table from Overview - NVIDIA Docs which shows that models were tested to run on DLA1+DLA2
and in Maximizing Deep Learning Performance on NVIDIA Jetson Orin with DLA | NVIDIA Technical Blog
which shows results on ruuning on GPU+DLA1+DLA2. How these results were obtained.
Additionally, can running INT8/FP16 model on DLA+GPU reduces the FPS comparing to running it on GPU?
These are the theoretical perf data when runs multiple pipelines on the different inferencing hardwares. E.G. One pipeline runs on GPU, one pipeline runs on DLA1 and the other pipeline runs on DLA2.
Thank you @Fiona.Chen
I tried to run custom DL on GPU+DLA which is in FP16 and when comparing the results with executing the same model in GPU only i found that running the model in DLA+GPU gives lower FPS compared to running it on GPU alone. Is there a specific justification for this, for example is it an issue because the model is in FP16 and if i generated the INT8 cali file this will help to increase FPS when running on DLA+GPU or is it an issue with the model it self?
When run one model on DLA+GPU, it may take more effort for transferring data between DLA and GPU and switch contexts for GPU and DLA frequently. It is better to run the whole model on DLA.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.