TensorFlow GPU not Working

Hello!

I use TensorFlow and Keras to train models on the Jetson Nano.
Here are some specs useful specs:
Python version: 3.6.9
TensorFlow version: 2.2.0
JetPack version: I don’t know the version or how to check what version I have. Downloaded it in late 2019. I can reinstall the latest JetPack if necessary to resolve the bugs…

I am getting errors whenever I import and train a model on the Nano.

Errors during the import statement:

2021-07-23 11:13:21.352365: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcudart.so.10.2'; dlerror: libcudart.so.10.2: cannot open shared object file: No such file or directory
2021-07-23 11:13:21.352446: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.

Errors during training:

2021-07-23 11:15:15.413643: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcuda.so.1
2021-07-23 11:15:15.487046: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:948] ARM64 does not support NUMA - returning NUMA node zero
2021-07-23 11:15:15.487277: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties: 
pciBusID: 0000:00:00.0 name: NVIDIA Tegra X1 computeCapability: 5.3
coreClock: 0.9216GHz coreCount: 1 deviceMemorySize: 3.87GiB deviceMemoryBandwidth: 194.55MiB/s
2021-07-23 11:15:15.487645: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcudart.so.10.2'; dlerror: libcudart.so.10.2: cannot open shared object file: No such file or directory
2021-07-23 11:15:15.488019: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcublas.so.10'; dlerror: libcublas.so.10: cannot open shared object file: No such file or directory
2021-07-23 11:15:15.488328: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcufft.so.10'; dlerror: libcufft.so.10: cannot open shared object file: No such file or directory
2021-07-23 11:15:15.488632: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcurand.so.10'; dlerror: libcurand.so.10: cannot open shared object file: No such file or directory
2021-07-23 11:15:15.488807: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcusolver.so.10'; dlerror: libcusolver.so.10: cannot open shared object file: No such file or directory
2021-07-23 11:15:15.488962: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcusparse.so.10'; dlerror: libcusparse.so.10: cannot open shared object file: No such file or directory
2021-07-23 11:15:15.489140: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcudnn.so.8'; dlerror: libcudnn.so.8: cannot open shared object file: No such file or directory
2021-07-23 11:15:15.489177: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1598] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2021-07-23 11:15:15.510076: W tensorflow/core/platform/profile_utils/cpu_utils.cc:106] Failed to find bogomips or clock in /proc/cpuinfo; cannot determine CPU frequency
2021-07-23 11:15:15.510954: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x3f5126f0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2021-07-23 11:15:15.511015: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2021-07-23 11:15:15.515619: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-07-23 11:15:15.515691: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1108]  

Similar issues have been encountered previously in the NVIDIA community, although the solutions provided are outdated (TensorFlow 1.x).
https://forums.developer.nvidia.com/t/tensorflow-gpu-not-working-in-nano/82171

I looked on the GPU support page on TensorFlow:
https://www.tensorflow.org/install/gpu
I don’t know how to install the given dependencies or check which ones are already installed.

Would really appreciate any inputs as to how I can train on the GPU to substantially speed up the training process. Feel free to ask me for code, versions of any libraries, etc. if necessary.

Thank you!

Hi,

Could not load dynamic library ‘libcudart.so.10.2’

Based on the error, TensorFlow cannot find the required CUDA version(10.2) in your Nano.
Please reflash your device with the latest JetPack version for it.

Thanks.

Hello!
Thank you so much. Just wanted a confirmation before flashing the latest image. Should doing this also resolve the other bugs?

Yes! I finished installing JetPack 4.5.1 along with TensorFlow and all associated dependencies. I ran the program and it worked flawlessly! It’s much faster now. Thank you so much!

Good to know this!
Thanks for the feedback.

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.