peterlee0127/tensorflow-nvJetson peterlee0127/tensorflow-nvJetson

  • Stars
    star
    202
  • Rank 193,691 (Top 4 %)
  • Language
    Python
  • Created over 7 years ago
  • Updated over 4 years ago
Twitter logo Share LinkedIn logo Share Facebook logo Share
peterlee0127/tensorflow-nvJetson
github

peterlee0127

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

TensorFlow for NVIDIA Jetson, also include patch and script for building.

tensorflow-nvJetson

TensorFlow for Nvidia Jetson TX1/TX2.

Install Latest Build of Tensorflow

Setup Environment

# Setting in .bashrc or .zshrc or other bash
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

$ sudo apt-get install libcupti-doc
export LD_LIBRARY_PATH=/usr/local/cuda/extras/CUPTI/lib64:$LD_LIBRARY_PATH

# Install python
$ sudo apt install python3 python3-dev

# Install hdf5
$ sudo apt-get install pkg-config libhdf5-100 libhdf5-dev

Install pip

$ wget https://bootstrap.pypa.io/get-pip.py -o get-pip.py
$ sudo python3 get-pip.py

Install numpy, keras

# requirements for numpy,keras
$ sudo apt install libhdf5-dev
$ sudo pip3 install numpy keras scipy

Install at Release

You can download wheel file at Release Page

Install by curl

sh -c "$(curl -fsSL https://tfjetson.peterlee0127.com/installTF.sh)"

Install by wget

sh -c "$(wget https://tfjetson.peterlee0127.com/installTF.sh -O -)"

This script will download lastest build tensorflow in this repository.

P.S. I recommend to donwload needed file, not use git clone. Using git clone will download all file in this repository.

Use the NVIDIA Official build

Python 2.7

pip install --extra-index-url=https://developer.download.nvidia.com/compute/redist/jp33 tensorflow-gpu

Python 3.5

pip3 install --extra-index-url=https://developer.download.nvidia.com/compute/redist/jp33 tensorflow-gpu

Nvidia Forum

TensorRT

Using TensorRT in TensorFlow

Install uff exporter for Jetson

TensorRT Test by TensorFlow

TensorRT test

Nvidia Jetson

JetPack 4.2, Tensorflow 2.0.0

2019 10/03

JetPack 3.3, TensorFlow 1.10

2018 8/13

  1. cuDNN v7.1.5
  2. CUDA 9.0
  3. Python 2.7 and Python 3.5
  4. TensorRT 4.0 GA

JetPack 3.2, TensorFlow 1.9

2018 7/11

  1. cuDNN 7.0
  2. CUDA 9.0
  3. Python 3.5

This package build with tensorRT.

JetPack 3.2, TensorFlow 1.8

2018 4/30.

  1. cuDNN 7.0
  2. CUDA 9.0
  3. Python 2.7

This package build with tensorRT.

JetPack 3.2, TensorFlow 1.7

2018 3/29.

  1. cuDNN 7.0
  2. CUDA 9.0
  3. Python 2.7

This package build with tensorRT.

JetPack 3.2, TensorFlow 1.6

  1. cuDNN 7.0
  2. CUDA 9.0
  3. Python 2.7

This package didn't build with tensorRT.

JetPack 3.2, TensorFlow 1.5

  1. cuDNN 7.0
  2. CUDA 9.0
  3. Python 2.7

If you had this kind of Memory Error.

2018-02-23 16:45:13.345534: W ./tensorflow/core/common_runtime/gpu/pool_allocator.h:195] could not allocate pinned host memory of size: 267264.  
2018-02-23 16:45:13.345585: E tensorflow/stream_executor/cuda/cuda_driver.cc:967] failed to alloc 240640 bytes on host: CUDA_ERROR_UNKNOWN.   
2018-02-23 16:45:13.345634: W ./tensorflow/core/common_runtime/gpu/pool_allocator.h:195] could not allocate pinned host memory of size: 240640.   
2018-02-23 16:45:13.345683: E tensorflow/stream_executor/cuda/cuda_driver.cc:967] failed to alloc 216576 bytes on host: CUDA_ERROR_UNKNOWN.

You can modify your tensorflow program. It should works.

config = tf.ConfigProto()
config.gpu_options.allow_growth = True

session = tf.Session(config=config, ...)

Install

Tensorflow 1.7.0
$ sudo pip install tensorflow-1.7.0-cp27-cp27mu-linux_aarch64.whl

Tensorflow 1.6.0
$ sudo pip install tensorflow-1.6.0-cp27-cp27mu-linux_aarch64.whl

Output of the test code

GPU Test

2017-07-26 17:21:02.457118: E tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:879] could not open file to read NUMA node: /sys/bus/pci/devices/0000:00:00.0/numa_node
Your kernel may have been built without NUMA support.
2017-07-26 17:21:02.457263: I tensorflow/core/common_runtime/gpu/gpu_device.cc:955] Found device 0 with properties:
name: NVIDIA Tegra X2
major: 6 minor: 2 memoryClockRate (GHz) 1.3005
pciBusID 0000:00:00.0
Total memory: 7.67GiB
Free memory: 5.30GiB
2017-07-26 17:21:02.457343: I tensorflow/core/common_runtime/gpu/gpu_device.cc:976] DMA: 0
2017-07-26 17:21:02.457374: I tensorflow/core/common_runtime/gpu/gpu_device.cc:986] 0:   Y
2017-07-26 17:21:02.457407: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1045] Creating TensorFlow device (/gpu:0) -> (device: 0, name: NVIDIA Tegra X2, pci bus id: 0000:00:00.0)
2017-07-26 17:21:02.457448: I tensorflow/core/common_runtime/gpu/gpu_device.cc:657] Could not identify NUMA node of /job:localhost/replica:0/task:0/gpu:0, defaulting to 0.  Your kernel may not have been built with NUMA support.
[[ 22.  28.]
 [ 49.  64.]]

test_tftrt.py

$ python test_tftrt.py
WARNING:tensorflow:From /usr/local/lib/python2.7/dist-packages/tensorflow/contrib/learn/python/learn/datasets/base.py:198: retry (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.
Instructions for updating:
Use the retry module or similar alternatives.
2018-04-02 11:25:15.649281: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:865] ARM64 does not support NUMA - returning NUMA node zero
2018-04-02 11:25:15.649495: I tensorflow/core/grappler/devices.cc:51] Number of eligible GPUs (core count >= 8): 0
2018-04-02 11:25:15.657161: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2624] Max batch size= 100 max workspace size= 33554432
2018-04-02 11:25:15.657245: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2630] starting build engine
2018-04-02 11:25:19.985906: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2635] Built network
2018-04-02 11:25:19.989301: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2640] Serialized engine
2018-04-02 11:25:19.990305: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2648] finished engine my_trt_op0 containing 7 nodes
2018-04-02 11:25:19.990493: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2668] Finished op preparation
2018-04-02 11:25:19.990663: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2676] OK finished op building
2018-04-02 11:25:20.027849: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1344] Found device 0 with properties:
name: NVIDIA Tegra X2 major: 6 minor: 2 memoryClockRate(GHz): 1.3005
pciBusID: 0000:00:00.0
totalMemory: 7.67GiB freeMemory: 1.83GiB
2018-04-02 11:25:20.027937: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
2018-04-02 11:25:20.027992: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-04-02 11:25:20.028024: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0
2018-04-02 11:25:20.028050: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:   N
2018-04-02 11:25:20.028165: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3926 MB memory) -> physical GPU (device: 0, name: NVIDIA Tegra X2, pci bus id: 0000:00:00.0, compute capability: 6.2)
2018-04-02 11:25:21.487230: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
2018-04-02 11:25:21.488576: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-04-02 11:25:21.488624: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0
2018-04-02 11:25:21.488659: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:   N
2018-04-02 11:25:21.488788: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3926 MB memory) -> physical GPU (device: 0, name: NVIDIA Tegra X2, pci bus id: 0000:00:00.0, compute capability: 6.2)
2018-04-02 11:25:21.570046: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
2018-04-02 11:25:21.570280: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-04-02 11:25:21.570316: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0
2018-04-02 11:25:21.570337: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:   N
2018-04-02 11:25:21.570446: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3926 MB memory) -> physical GPU (device: 0, name: NVIDIA Tegra X2, pci bus id: 0000:00:00.0, compute capability: 6.2)
2018-04-02 11:25:21.628937: I tensorflow/core/grappler/devices.cc:51] Number of eligible GPUs (core count >= 8): 0
2018-04-02 11:25:21.635393: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2624] Max batch size= 100 max workspace size= 33554432
2018-04-02 11:25:21.635480: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2628] Using FP16 precision mode
2018-04-02 11:25:21.635507: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2630] starting build engine
2018-04-02 11:25:22.054581: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2635] Built network
2018-04-02 11:25:22.056254: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2640] Serialized engine
2018-04-02 11:25:22.056768: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2648] finished engine my_trt_op1 containing 7 nodes
2018-04-02 11:25:22.056962: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2668] Finished op preparation
2018-04-02 11:25:22.057143: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2676] OK finished op building
2018-04-02 11:25:22.075579: I tensorflow/core/grappler/devices.cc:51] Number of eligible GPUs (core count >= 8): 0
2018-04-02 11:25:22.081608: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2410] finished op preparation
2018-04-02 11:25:22.081704: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2418] OK
2018-04-02 11:25:22.081732: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2419] finished op building
2018-04-02 11:25:22.112265: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
2018-04-02 11:25:22.112386: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-04-02 11:25:22.112424: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0
2018-04-02 11:25:22.112452: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:   N
2018-04-02 11:25:22.112562: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3926 MB memory) -> physical GPU (device: 0, name: NVIDIA Tegra X2, pci bus id: 0000:00:00.0, compute capability: 6.2)
2018-04-02 11:25:22.199192: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
2018-04-02 11:25:22.199323: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-04-02 11:25:22.199350: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0
2018-04-02 11:25:22.199375: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:   N
2018-04-02 11:25:22.199478: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3926 MB memory) -> physical GPU (device: 0, name: NVIDIA Tegra X2, pci bus id: 0000:00:00.0, compute capability: 6.2)
2018-04-02 11:25:22.239846: W tensorflow/contrib/tensorrt/log/trt_logger.cc:34] DefaultLogger Int8 support requested on hardware without native Int8 support, performance will be negatively affected.
2018-04-02 11:25:22.626763: I tensorflow/contrib/tensorrt/convert/convert_graph.cc:298] Starting Calib Conversion
2018-04-02 11:25:22.627250: I tensorflow/contrib/tensorrt/convert/convert_graph.cc:310] Num Calib nodes in graph= 1
2018-04-02 11:25:23.703319: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1423] Adding visible gpu devices: 0
2018-04-02 11:25:23.703421: I tensorflow/core/common_runtime/gpu/gpu_device.cc:911] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-04-02 11:25:23.703452: I tensorflow/core/common_runtime/gpu/gpu_device.cc:917]      0
2018-04-02 11:25:23.703475: I tensorflow/core/common_runtime/gpu/gpu_device.cc:930] 0:   N
2018-04-02 11:25:23.703567: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1041] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3926 MB memory) -> physical GPU (device: 0, name: NVIDIA Tegra X2, pci bus id: 0000:00:00.0, compute capability: 6.2)
Pass
Tensorflow 1.7(build with TensorRT) is larger than 100MB. I split the whl file to 2 part. Please use following command to merge file.

merge file

$ cat tensorflow-1.7.0-cp27-cp27mu-linux_aarch64.whl.part-* > tensorflow-1.7.0-cp27-cp27mu-linux_aarch64.whl

split file

$ split -b 70m tensorflow-1.7.0-cp27-cp27mu-linux_aarch64.whl tensorflow-1.7.0-cp27-cp27mu-linux_aarch64.whl-part-

Install System on SSD (Solid State Disk)

You can find information at jetsonhacks.

jetsonhacks-install-samsung-ssd-on-nvidia-jetson-tx1

buymeacoffee

More Repositories

1

RealTimeSubtitle

A plug-in show real time Subtitle/Text on your live stream (like OBS studio).
JavaScript
31
star
2

DNS-Security-iOS

The First iOS app with DNS over HTTPS and DNS over TLS in iOS 14. http://dns-security.peterlee.app/getapp
Ruby
10
star
3

iBeacon_AutoCheckIn

User Auto CheckIn using iBeacon BLE technology
Objective-C
9
star
4

Streaming-Pipeline

A real-time text classification based on Kafka and Spark.
Scala
7
star
5

YeeLight-Blue-2

YeeLight Blue 2
JavaScript
7
star
6

GameController_MFI

GameController_MFI iOS 7
Objective-C
5
star
7

AirBox

An open-source pm2.5 sensor, website and ELK.
C++
4
star
8

Walker-Analytics

Walker-Analytics
Swift
2
star
9

IP-Tool

Display WiFi and Cellular IP Address on your iPhone (without Jailbreak)
Objective-C
2
star
10

Geofence-iOS

Geofence for iOS ,Region Monitoring work in App Closed ,Open, Background.
Objective-C
2
star
11

homekit-neopixel-rpi

homekit-neopixel-rpi. work with https://www.studiopieters.nl/homebridge-neopixel-light/
Python
1
star
12

EWS-Calendar

Sync Microsoft Outlook Calendar to Google Calendar and generate web api
JavaScript
1
star
13

AVSpeechSynthesizer

speech multiple Language on your iOS Device , required iOS 7
Objective-C
1
star
14

csvToGoogleSheets

A tool for uploading csv to google sheets.
JavaScript
1
star
15

OSX-Maintenance-Script

OSX Maintenance Script, clear unused account , setting ,cache
Shell
1
star
16

RocketChat-for-Sandstorm

new RocketChat for Sandstorm
1
star