Is there a upgrade for TensorFlow for Jetson TX2 JP3.3?

I followed the instruction here for tf-gpu but could not install tf-gpu for python 3.7. Error: not found a version that satisfies the requirement

I need tf-gpu for python v3.7 or 3.8 or worst case 3.6. This is because the lambda function in AWS only seem to support 3.6-3.8 now. I need it for greengrass

Currently i need {Python 3.7+JetPack3.3} or worst case {Python 3.6+JetPack3.3}

Is this available?

Do i need to upgrade my jetpack in this case? If so, how do i do it?


We only provide the package with default python version.

JetPack3.3 + python3.5 (Ubuntu16.04)

sudo pip3 install --pre --extra-index-url tensorflow-gpu==1.14.0+nv19.7

JetPack4.x + python3.6 (Ubuntu18.04)

sudo pip3 install --pre --extra-index-url tensorflow-gpu==1.14.0+nv19.10

If none of above package can fit your requirement, you can also try to build TensorFlow from source directly.
Here is a recommended steps for your reference:


is there a reason why pip/pip3 install tensorflow-gpu is not possible?

It seems too much of a hassle to build one on my own. I guess i will continue to use python2 for now. Is the team planning to build a JetPack3.3 + python3.6 tf-gpu in the coming year?

Python2 will no longer be supported in the coming year so aws greengrass might remove python2 from their lambda funcion. New greengrass users will not be able to use tf-gpu on Tx2 during that period.

Also is there a way to build to upgrade my TX2 jetpack 3.3 to 4.2/latest?


JetPack3.3 is using Ubuntu16.04, which by default installed with python3.5.
So you won’t get a JetPack3+python3.6, unless you build on your own.

To get a python3.6 package, please upgrade your device with JetPack4.x.
TensorFlow doesn’t officially support ARM system so there is no prebuilt package available.



Is it easy and safe to upgrade to 4.x. Are there any clear instruction somewhere on how to do it?