Can nvidia-tensorflow (1.x) be used with RTX 4090



I understand the above mentioned tensorflow can be used with RTX30 series GPU.

Can this tensorflow be used with RTX4090?

It is possible for NVIDIA version of TF 1.15.x to run on a 4090.

The only recommendation I have would be to use NGC.
The latest NGC container supports TF 1.15.5 and CUDA 12. That will work on a 4090.

I won’t be able to give you detailed instructions for usage. There is a NGC section on these forums.

Furthermore, if you don’t want to use NGC, I won’t be able to give you a detailed build recipe. Nevertheless it should be possible to build TF 1.15.x for that target, and the NGC container is proof.

Thank you for this answer.
It is clear and provides solid ideas to move forward with.

I may try NGC on 3090 I have access to and at least get that going before I build and buy the rig with the 4090.

Does anyone know if I can I rent a 4090 in the cloud at an hourly rate to try it?

How is it going? Are you able to run NGC on 3090 for tensorflow 1.x?

Confirmed running 4090 and 3060 with TF 1.15.5.
(NGC 23.01-tf1-py3)

I’ve tried so many times with tensorflow official and other docker images but failed.
But NGC images works without any problem.
Thank you Robert_Crovella for saving my time.

1 Like