Compatibility between Jetpack 4.6.1, cuda 10.2 and Tensorflow

Hi all,
I have a Jetson AGX Xavier with jetpack 4.6.1, the pre installed cuda with this jetpack is 10.2

I want now to install tensorflow from source.
The required TF for my occasion according to this documentation is TF ==2.7.0 +nv22.01

But as you can see TF=2.7 needs cuda 11.2

I tried to install manually TF=2.3.0 but i get this message:

“ERROR: Could not find a version that satisfies the requirement tensorflow==2.3.0+nv20.9 (from versions: 1.15.5+nv22.1, 2.7.0+nv22.1)
ERROR: No matching distribution found for tensorflow==2.3.0+nv20.9”

Is it necessary to downgrade Jetpack or to upgrade my cuda version.
By upgrading cuda may this arise any problem?

Is there any other solution?

Tnx

Hi,

You can install v2.7.0+nv22.1 on JetPack 4.6.1 directly.
The compatibility is solved and tested by our internal team.

Thanks.

Hi,

Of course, the first thing I did was to install TF v2.7 but when I run my module, I get this message :

“2022-06-07 15:28:06.214347: W tensorflow/stream_executor/gpu/asm_compiler.cc:111] ** WARNING ** You are using ptxas 10.2.300, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.”

It’s obvious a warning msg about the compatibility between Cuda and TF

So I try to figure out a solution among all those warnings

Thanks

Hi,

Could you share a simple reproducible script for the error?
We test the TF package on a Xavier board and it can run correctly.

$ python3
Python 3.6.9 (default, Mar 15 2022, 13:55:28) 
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
>>> print(tf.config.experimental.list_physical_devices('GPU'))
2022-06-13 15:41:00.615857: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-13 15:41:00.736515: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-13 15:41:00.737156: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
>>>

Thanks.

Hi,

In order to proceed with my work, I downgraded the Jetpack to 4.5.1.

When I finish my task, I will answer again to this topic

Thanks

Is this still an issue to support? Any result can be shared? Thanks

Hi, no I work now on jetpack 4.5.1 which works fine by my side.

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