Introduction (Abstract)

Jetson Nano JETPACK 44.1 has installed cuda Library 10.2. And I tried to install tensorflow with 'python3 -m pip', but it was not working properly. Finally, I can make tensowflow working with gpu by other install index.

Problem

I download the SD card image file from Nvidia Jetson Download Center and follow the instruction and installed L4T 44.1 on my Jetson Nano.

Then, I tried install tensorflow for python:

$ python3 -m pip install --upgrade pip
$ python3 -m pip install tensorflow-gpu

Next, I opend python3 interactive mode, try import tensorflow:

$ python3
Python 3.6.9 (default, Oct  8 2020, 12:12:24) 
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from tensorflow.python.client import device_lib

But, it did not finish successfully:

Traceback (most recent call last):
File "/home/tri/.local/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow.py", line 58, in <module>
from tensorflow.python.pywrap_tensorflow_internal import *
File "/home/tri/.local/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in <module>
_pywrap_tensorflow_internal = swig_import_helper()
File "/home/tri/.local/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
_mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
File "/usr/lib/python3.6/imp.py", line 243, in load_module
return load_dynamic(name, filename, file)
File "/usr/lib/python3.6/imp.py", line 343, in load_dynamic
return _load(spec)
ImportError: libcublas.so.10.0: cannot open shared object file: No such file or directory

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/tri/.local/lib/python3.6/site-packages/tensorflow/__init__.py", line 24, in <module>
from tensorflow.python import pywrap_tensorflow  # pylint: disable=unused-import
File "/home/tri/.local/lib/python3.6/site-packages/tensorflow/python/__init__.py", line 49, in <module>
from tensorflow.python import pywrap_tensorflow
File "/home/tri/.local/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow.py", line 74, in <module>
raise ImportError(msg)
ImportError: Traceback (most recent call last):
File "/home/tri/.local/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow.py", line 58, in <module>
from tensorflow.python.pywrap_tensorflow_internal import *
File "/home/tri/.local/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in <module>
_pywrap_tensorflow_internal = swig_import_helper()
File "/home/tri/.local/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
_mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
File "/usr/lib/python3.6/imp.py", line 243, in load_module
return load_dynamic(name, filename, file)
File "/usr/lib/python3.6/imp.py", line 343, in load_dynamic
return _load(spec)
ImportError: libcublas.so.10.0: cannot open shared object file: No such file or directory


Failed to load the native TensorFlow runtime.

See https://www.tensorflow.org/install/errors

for some common reasons and solutions.  Include the entire stack trace
above this error message when asking for help.

I thought that the problem occured by this tensorflow version required the cuda 10.0, but L4T 44.1 has cuda 10.2, the reference version is different. I investigated it because I thought it was because it tried to refer to the illegal version.

Investigation

I found a site which is written about Jetson Nano and Tensowflow installation.
In this site information, the 4th Item, the author wrote about the dedicated version for Jetson Nano. So I accessed that site and found v44 directory. But, I found some Tensorflow~.whl files. Which is the best one? Check [the document about JetPack SDK] (https://developer.nvidia.com/embedded/jetpack), I decided to use Tensorflow 2.x. and to try the newest version.

$ python3 -m pip install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v44 tensorflow==2.3.1

Then, try with Python3:

$ python3
Python 3.6.9 (default, Oct  8 2020, 12:12:24) 
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow
2020-12-28 17:02:48.297018: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.2

Yeah! I got a success.

And try to list_local_devices():

>>> from tensorflow.python.client import device_lib
>>> device_lib.list_local_devices()
2020-12-28 17:04:10.089230: W tensorflow/core/platform/profile_utils/cpu_utils.cc:108] Failed to find bogomips or clock in /proc/cpuinfo; cannot determine CPU frequency
2020-12-28 17:04:10.090292: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x37c6caf0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-12-28 17:04:10.090373: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-12-28 17:04:10.112908: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcuda.so.1
2020-12-28 17:04:10.233309: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1046] ARM64 does not support NUMA - returning NUMA node zero
2020-12-28 17:04:10.233678: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x36c08980 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-12-28 17:04:10.233746: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): NVIDIA Tegra X1, Compute Capability 5.3
2020-12-28 17:04:10.235468: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1046] ARM64 does not support NUMA - returning NUMA node zero
2020-12-28 17:04:10.236124: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1742] Found device 0 with properties: 
pciBusID: 0000:00:00.0 name: NVIDIA Tegra X1 computeCapability: 5.3
coreClock: 0.9216GHz coreCount: 1 deviceMemorySize: 3.86GiB deviceMemoryBandwidth: 194.55MiB/s
2020-12-28 17:04:10.236236: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.2
2020-12-28 17:04:10.253989: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublas.so.10
2020-12-28 17:04:10.271843: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcufft.so.10
2020-12-28 17:04:10.278697: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcurand.so.10
2020-12-28 17:04:10.297505: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusolver.so.10
2020-12-28 17:04:10.309289: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusparse.so.10
2020-12-28 17:04:10.311692: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudnn.so.8
2020-12-28 17:04:10.311971: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1046] ARM64 does not support NUMA - returning NUMA node zero
2020-12-28 17:04:10.312283: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1046] ARM64 does not support NUMA - returning NUMA node zero
2020-12-28 17:04:10.312368: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1884] Adding visible gpu devices: 0
2020-12-28 17:04:10.313306: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.2
2020-12-28 17:04:13.541741: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1283] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-12-28 17:04:13.542731: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1289]      0 
2020-12-28 17:04:13.542820: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1302] 0:   N 
2020-12-28 17:04:13.549035: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1046] ARM64 does not support NUMA - returning NUMA node zero
2020-12-28 17:04:13.550273: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1046] ARM64 does not support NUMA - returning NUMA node zero
2020-12-28 17:04:13.552152: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1428] Created TensorFlow device (/device:GPU:0 with 61 MB memory) -> physical GPU (device: 0, name: NVIDIA Tegra X1, pci bus id: 0000:00:00.0, compute capability: 5.3)
[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 5071338378085176143
, name: "/device:XLA_CPU:0"
device_type: "XLA_CPU"
memory_limit: 17179869184
locality {
}
incarnation: 5701248517285533579
physical_device_desc: "device: XLA_CPU device"
, name: "/device:XLA_GPU:0"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 3113596419989883465
physical_device_desc: "device: XLA_GPU device"
, name: "/device:GPU:0"
device_type: "GPU"
memory_limit: 64991232
locality {
  bus_id: 1
  links {
  }
}
incarnation: 10370280253443219495
physical_device_desc: "device: 0, name: NVIDIA Tegra X1, pci bus id: 0000:00:00.0, compute capability: 5.3"
]

Conclusion

Tensorflow installation on Jetson Nano L4T 44.1 (2020/10/21) (For Python3.6.9)

$ python3 -m pip install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v44 tensorflow==2.3.1