[Install] On Jetson nano orin

[truonghau8]

问题描述 Issue Description

Hello every one,

I try to install paddle paddle on Jetson Nano Orin.

Currently I using jetpack 6.x. But I cannot see any file release like *.whl (support gpu on jetson cuda 12.6).
Does any one have any solution for that ? or try to re-build compile paddle paddle with Jetson nano orin (Jetpack 6.x)

Thank you

版本&环境信息 Version & Environment Information

Paddle Version: 3.0
GPU: Cuda 12.6
OS: Jetpack 6.x

Jetson nano orin

[quangsonle]

same problem, i dont even find the whl files for ARM architecture at all. these are all whl files i found from paddle page: https://www.paddlepaddle.org.cn/packages/stable/cu116/paddlepaddle-gpu/ and no ARM version (cu118 of jetpack5., you can check jetpack 6. with cuda 12 here:https://www.paddlepaddle.org.cn/packages/stable/cu126/paddlepaddle-gpu/ also no ARM version)

i tried to build Paddle 3.1 from source on jetson and failed of course

[risemeup1]

paddle暂时不支持jetson

[EmmonsCurse]

@truonghau8 You can try compiling it yourself using the source code build method.

git clone https://github.com/PaddlePaddle/Paddle.git
cd Paddle

mkdir -p build && cd ./build

cmake .. -DCMAKE_CXX_FLAGS='-Wno-error -w'  -DWITH_CONTRIB=OFF   -DWITH_MKL=OFF    -DWITH_MKLDNN=OFF   -DWITH_TESTING=OFF   -DCMAKE_BUILD_TYPE=Release   -DON_INFER=ON   -DWITH_PYTHON=ON -DPY_VERSION=3.10  -DWITH_XBYAK=OFF  -DWITH_NV_JETSON=ON -DWITH_TENSORRT=ON -DTENSORRT_ROOT=/usr -DCMAKE_CUDA_COMPILER=/usr/local/cuda-11.6/bin/nvcc  -DWITH_NCCL=OFF -DCUDA_ARCH_NAME=Auto -DREDUCE_INFERENCE_LIB_SIZE=ON -DPYTHON_EXECUTABLE=/usr/bin/python -DWITH_SHARED_PHI=ON -DWITH_OPENVINO=OFF

ulimit -n 65535 && make -j 6

[fangfangssj]

@truonghau8 hi, I compiled a Paddle on the official image of nvidia jetson jetpack 6.2.1 and you can see if it works
通过网盘分享的文件：paddlepaddle_gpu-0.0.0-cp310-cp310-linux_aarch64.whl
链接: https://pan.baidu.com/s/1ngFAMd7udXLzZNLPT6zdyg 提取码: 7q9q

版本&环境信息 Version & Environment Information

Python Version: 3.10
Paddle Version: dev20250720
GPU: Cuda 12.6
OS: Jetpack 6.2.1
Device: Jetson orin nano

[quangsonle]

@fangfangssj do you have the c++ inference version of jetpack 6.2.1?

[fangfangssj]

@fangfangssj do you have the c++ inference version of jetpack 6.2.1?

Sorry, I didn't compile.

[truonghau8]

@truonghau8 hi, I compiled a Paddle on the official image of nvidia jetson jetpack 6.2.1 and you can see if it works 通过网盘分享的文件：paddlepaddle_gpu-0.0.0-cp310-cp310-linux_aarch64.whl 链接: https://pan.baidu.com/s/1ngFAMd7udXLzZNLPT6zdyg 提取码: 7q9q

版本&环境信息 Version & Environment Information

Python Version: 3.10 Paddle Version: dev20250720 GPU: Cuda 12.6 OS: Jetpack 6.2.1 Device: Jetson orin nano

Hello fangfangssj,

I am using your whl for python. It's working fine. Many thank from your whl.

Could you give me the build step of your compile? Like cmake command, ...

[fangfangssj]

@truonghau8 hi, I compiled a Paddle on the official image of nvidia jetson jetpack 6.2.1 and you can see if it works 通过网盘分享的文件：paddlepaddle_gpu-0.0.0-cp310-cp310-linux_aarch64.whl 链接: https://pan.baidu.com/s/1ngFAMd7udXLzZNLPT6zdyg 提取码: 7q9q

版本&环境信息 Version & Environment Information

Python Version: 3.10 Paddle Version: dev20250720 GPU: Cuda 12.6 OS: Jetpack 6.2.1 Device: Jetson orin nano

Hello fangfangssj,

I am using your whl for python. It's working fine. Many thank from your whl.

Could you give me the build step of your compile? Like cmake command, ...

Here's the command I compiled.

git clone https://github.com/PaddlePaddle/Paddle.git
cd Paddle

mkdir -p build && cd ./build

cmake .. -DPY_VERSION=3.10 \
         -DWITH_MKL=OFF \
         -DWITH_TESTING=OFF \
         -DCMAKE_BUILD_TYPE=Release \
         -DON_INFER=ON \
         -DWITH_PYTHON=ON \
         -DWITH_XBYAK=OFF \
         -DWITH_NV_JETSON=ON \
         -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda \
         -DWITH_NCCL=OFF \
         -DWITH_RCCL=OFF \
         -DWITH_DISTRIBUTE=OFF \
         -DWITH_GPU=ON \
         -DWITH_TENSORRT=ON \
         -DWITH_ARM=ON

ulimit -n 65535 && make TARGET=ARMV8 -j3

pip install python/dist/paddlepaddle_gpu-0.0.0-cp310-cp310-linux_aarch64.whl

[ibnbd]

I tried to build from the source, and the CMake version was having issues.

[fangfangssj]

I tried to build from the source, and the CMake version was having issues.

My CMake version is 3.22.1, and Paddle currently does not support CMake version 4.0 and above.

[quangsonle]

any of you tried trt_fp16 quantization on jetson? in my case it works on AMD but on jetson it always return no detection (while "paddle" mode works on jetson) even with the build from source library by @fangfangssj 's suggestion