FPS performance drop issue on AGX Orin

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) Jetson AGX Orin 32GB
• DeepStream Version DS6.1
• JetPack Version (valid for Jetson only)
• TensorRT Version
• NVIDIA GPU Driver Version (valid for GPU only)
• Issue Type( questions, new requirements, bugs) questions

Hi,

We have a use case with 3 gst pipelines running simultaneously that involves the following encoding & decoding activities.
Encode: x1 1080p60 + x1 1080p30 + x1 4Kp60
Decoding: x12 1080p30 + x2 4Kp30

We collect the FPS performance for this use case using H264 video as input as well as H265 video .
When we run this use case on AGX Xavier (32GB), the 3 gst pipelines able to meet the target FPS (30 fps /60 fps) with both H264 and H265 video input. However, when we run the same use case on AGX Orin (32GB), the FPS drops below expected target (~27 FPS / ~53FPS) for H264 video input while the FPS performance with H265 video still able to meet the target FPS. It seems to us that this is due to the encoder/decoder HW limitation on Orin as we see the encoder/decoder capabilities as published in the link below shows that the the encoder/decoder on AGX Orin has lower performance than AGX Xavier. With this encoder/decoder limitation on AGX Orin, is there any suggestion on how we can workaround this issue so that our use case can achieve the same FPS performance on AGX Orin as well?

https://developer.nvidia.com/embedded/jetson-modules

5

What is the video resolution?

6

These are the high level view of the 3 pipelines running simultaneously:

(Note: 1080p means 1920x1080, 4Kp means 3840x2160)

Pipeline 1
x12 1080p30 video inputs → x12 nvv4l2decoder → batched with streammux → tiled as 3x4 window → display on 4K monitor

Pipeline2
x1 4Kp30 video input → nvv4l2decoder-> nvinfer → fakesink

Pipeline3 (note: “|->” are the subsequent pipelines linked to the tee)
x1 4Kp30 video input → nvv4l2decoder → tee
|-> downscale to 1080p > encode > encrypt> filesink (1080p30)
|-> upframe 60fps >downscale to 1080p > encode > encrypt >filesink(1080p60)
|->upframe 60fps > encode > encrypt > filesink (4Kp60)

7

It exceeds the HW limitation of Orin. I think you have read the spec. Jetson Modules | NVIDIA Developer

8

"
We have a use case with 3 gst pipelines running simultaneously that involves the following encoding & decoding activities.
Encode: x1 1080p60 + x1 1080p30 + x1 4Kp60
Decoding: x12 1080p30 + x2 4Kp30
"
Mixture of 1080p and 4k

9

We expected Orin’s codec Accl. to perform better than Xavier not worse.

10

We will consider better codec capability in future product.

11

Despite of PVA and VIA in Orin, can we conclude that Orin has lesser VIA HW capability than Xavier?

12

The HW capability is as declared in the spec: Jetson Modules | NVIDIA Developer

14

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.
Thanks

15

The CPU/HW encoder/decoder capabilities as advertised for AGX Orin DevKit here https://developer.nvidia.com/embedded/jetson-agx-orin-developer-kit is different from the one shown in https://developer.nvidia.com/embedded/jetson-modules. In this case, does this means the module on the AGX Orin devkit is not equivalent to the AGX Orin 32GB module? Also, for AGX Orin devkit what’s the HW encoder/decoder capabilities support for H264 video? The page here https://developer.nvidia.com/embedded/jetson-agx-orin-developer-kit only advertise the capability for H265 file

16

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.