JetPack 3.1 post installation on TX2 issue

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

I have test caffe SSD model on TX2 platform,but only got 3 fps.From other issues’ suggestion,JetPack 3.1 which contains TensorRT2.1 and cuDNN v6 can speed up 2x.
As I post install TesnsorRT and cuDNN on target,there is problems like this:
Get:1 file:/var/cuda-repo-8-0-local InRelease
Ign:1 file:/var/cuda-repo-8-0-local InRelease
Get:2 file:/var/libopencv4tegra-repo InRelease
Ign:2 file:/var/libopencv4tegra-repo InRelease
Get:3 file:/var/nv-gie-repo-ga-cuda8.0-gie1.0-20170116 InRelease
Ign:3 file:/var/nv-gie-repo-ga-cuda8.0-gie1.0-20170116 InRelease
Get:4 file:/var/nv-gie-repo-ga-cuda8.0-trt2.1-20170614 InRelease
Ign:4 file:/var/nv-gie-repo-ga-cuda8.0-trt2.1-20170614 InRelease
Get:5 file:/var/visionworks-repo InRelease
Ign:5 file:/var/visionworks-repo InRelease
Get:6 file:/var/visionworks-sfm-repo InRelease
Ign:6 file:/var/visionworks-sfm-repo InRelease
Get:7 file:/var/visionworks-tracking-repo InRelease
Ign:7 file:/var/visionworks-tracking-repo InRelease
Get:8 file:/var/cuda-repo-8-0-local Release [574 B]
Get:9 file:/var/libopencv4tegra-repo Release [347 B]
Get:10 file:/var/nv-gie-repo-ga-cuda8.0-gie1.0-20170116 Release [574 B]
Get:11 file:/var/nv-gie-repo-ga-cuda8.0-trt2.1-20170614 Release [574 B]
Get:8 file:/var/cuda-repo-8-0-local Release [574 B]
Get:12 file:/var/visionworks-repo Release [1,999 B]
Get:9 file:/var/libopencv4tegra-repo Release [347 B]
Get:13 file:/var/visionworks-sfm-repo Release [2,003 B]
Get:10 file:/var/nv-gie-repo-ga-cuda8.0-gie1.0-20170116 Release [574 B]
Get:14 file:/var/visionworks-tracking-repo Release [2,008 B]
Get:11 file:/var/nv-gie-repo-ga-cuda8.0-trt2.1-20170614 Release [574 B]
Get:12 file:/var/visionworks-repo Release [1,999 B]
Get:13 file:/var/visionworks-sfm-repo Release [2,003 B]
Get:14 file:/var/visionworks-tracking-repo Release [2,008 B]
0% [Working]

thanks, how can i fix this?

2

it seems network problems.when I do this command "sudo apt-get update "on TX2,meet the same problem.

3

Hi,

Please maximize device performance first:

sudo nvpmodel -m 0
sudo ~/jetson_clocks.sh
4

thanks ,AastaLLL.
I had maximize device to model 0.

5

Hi,

What framerate do you expect?

We also have some users use SSD on TX2. They got similar framerate and planned to accelerate it with TensorRT.
Here for your reference:
[url]https://devtalk.nvidia.com/default/topic/1007313/how-to-build-the-objection-detection-framework-ssd-with-tensorrt-on-tx2-/[/url]

6

Thanks,AastaLLL.
From this reference, i know you had got 8~9 fps (SSD model)on TX2 using Jetpack 3.1.
https://devtalk.nvidia.com/default/topic/1021809/jetson-tx2/can-the-nvidia-tensorrt-accelerate-ssd-single-shot-detector-/1

So, today I flash my tx2 platform using jetpack 3.1, it speed up 2x as expected up to 5~8 fps.But the fps flows up and down,and i do the command “sudo ~/tegrastats”,shows the GR3D GPU %0,%99,%57,and etc.The frequency also fluctuates up and down.
Can I get a fixed processing FPS?

7

Please notice that fps=8-9 is from ssd_pascal_video.py.

For web-camera, it’s recommended to use TensorRT instead of Caffe to get better performance.

8

Thanks,AastaLLL.
I do run ssd_pascal_video.py on tx2, and use the caffemodel download from github.How can I speed up further more using caffe? And is it normal while the FPS changes wide scope?

9

Hi,AastaLLL.
I think use FP16 instead of FP32 can speed up. How can I change SSD caffe model to FP16,can you give some suggestions?

10

Hi,

Sorry for the late reply.

Please make sure you are on JetPack3.1 and maximized the CPU/GPU frequency.
Usually, we can get a stable 8-9 fps results.

For fp16 issue:
Since SSD is a particular Caffe branch, please check the source if it supports fp16 mode.
Thanks.