Tensorflow v2.16.1 GPU version local build on Jetson Orin Nano failed

Tensorflow v2.16.1 GPU version local build on Jetson Orin Nano board, failed with following error.

Any idea?

ERROR: /home/daniel/.cache/bazel/_bazel_daniel/11588ef030db288b35dd97b3c9d34cbc/external/local_xla/xla/stream_executor/cuda/BUILD:505:13: Compiling xla/stream_executor/cuda/cuda_conditional_kernels.cu.cc failed: (Exit 1): clang failed: error executing command (from target @local_xla//xla/stream_executor/cuda:cuda_conditional_kernels) /usr/lib/llvm-14/bin/clang -MD -MF bazel-out/aarch64-opt/bin/external/local_xla/xla/stream_executor/cuda/_objs/cuda_conditional_kernels/cuda_conditional_kernels.cu.pic.d ... (remaining 72 arguments skipped)
clang: warning: CUDA version is newer than the latest supported version 11.5 [-Wunknown-cuda-version]
clang: error: unsupported CUDA gpu architecture: sm_90
Target //tensorflow/tools/pip_package:build_pip_package failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 853.288s, Critical Path: 61.65s
INFO: 11696 processes: 8427 internal, 3269 local.
FAILED: Build did NOT complete successfully
Package: nvidia-jetpack
Version: 6.0-b52
Priority: standard
Section: metapackages
Maintainer: NVIDIA Corporation
Installed-Size: 199 kB
Depends: nvidia-jetpack-runtime (= 6.0-b52), nvidia-jetpack-dev (= 6.0-b52)
Homepage: http://developer.nvidia.com/jetson
Download-Size: 29.3 kB
APT-Manual-Installed: yes
APT-Sources: https://repo.download.nvidia.com/jetson/common r36.2/main arm64 Packages
Description: NVIDIA Jetpack Meta Package

Hi

Do you need TensorFlow v2.16 or want a CUDA version of TensorFlow?
Please follow the command in the comment to get the CUDA TensorFlow package:

Thanks.

I have found there is no 2.16.1 from nvidia repo, which can be simply use

$ sudo pip3 install --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v60dp tensorflow==2.16.1+nv....

So I just wanna build this version, and met this compile error.

Is there any way to build from source?

EDIT: This is a pic from NVIDIA binary repo, so 2.16.1 is not supported right now.

Hi,

2.16 is not available yet.
You can find some building steps on the TensorFlow page.

As mentioned in topic 291321, please check with the TensorFlow team if the steps are not working.

Thanks.

Currently, we have found that the same code from tensorflow document. Running on Jetson Orin Nano , the result is quite un-beleivable. We don’t know why.

Here is the comparison of NVIDIA 2.15.0+nv24.03 v.s. Colab v.s. Tensorflow Documentation.

I DO think much more attention should be take care of thoes warnings. So we wanna know how cross compile NVIDIA did. Are thoese warnings correct??? How to compile tensorflow for Jetson Orin Nano?

Tensorflow issue:

Hi,

Unfortunately, we don’t have a public document about how we build the TensorFlow package.
But it should be very close to the instructions shared on the TensorFlow document.

If you have concerns about the functionality of our prebuilt, we can keep discussing it on the other topic you file:

Thanks.

Thanks, I suggest use this ticket to track the document of how to build tensorflow on Jetson Orin Nano.

And I also create a new topic on the inconsistency of tensorflow demo code: Inconsistency of NVIDIA 2.15.0+nv24.03 v.s. Colab v.s. Tensorflow Documentation

Hi,

We don’t have a document for building TensorFlow since it’s a third-party library.
In general, we follow the same steps as the TensorFlow document.

Thanks.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.