Is an update from TF=1.8.0-gpu-py3 to 1.15.0-gpu-py3 in the pipeline

[santosh-shriyan]

The reason for this request is that newer hardware like the rtx30 series do not support cuda10.1 for TF=1.8.0.
I have managed to train a model on 1.15.0-gpu-py3 docker image but the train model do not inference with the current code base.

[tom-doerr]

It's not planned. Could it help to try the versions between 1.8 and 1.15?

…

On Thu, Jun 24, 2021, 16:05 Santosh Shriyan ***@***.***> wrote: The reason for this request is that newer hardware like the rtx30 series do not support cuda10.1 for TF=1.8.0. I have managed to train a model on 1.15.0-gpu-py3 docker image but the train model do not inference with the current code base. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#11>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFSYSFFXA7GFUYTMFQ4KNPDTUM3TVANCNFSM47H64SYQ> .

[santosh-shriyan]

I am already trying it out... unfortunately the lowest rtx 30 series goes to is 460 driver which is cuda 11.2.
That becomes a problem.