Building Tensorflow from source with docker: how to change cuda version?

rajao · November 5, 2021, 9:27am

Dear,

We would like to rebuild tensorflow from source on your compute cluster to take advantage of hardware optimisation. We use the build based on docker since it’s difficult to install build dependencies on this cluster.
The build using cuda default option works but are relatively slow at runtime. We’re wondering if it is due to the fact that the build was done with cuda 11.2 and cuDNN 8 and in the cluster with installed cuda 11.4 ?
The other question is how to change the cuda and cudnn version when building tensorflow from sources ?

Best,

Jaonary

Bhack · November 5, 2021, 12:08pm

I think that you can upvote and subscribe to:

E.g. the RBE image used in the CI was updated:

But not in the tensorflow/docker-deve GPU:

github.com

angerson/tensorflow/blob/master/tensorflow/tools/dockerfiles/dockerfiles/devel-gpu.Dockerfile

# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
#
# THIS IS A GENERATED DOCKERFILE.
#
# This file was assembled from multiple pieces, whose use is documented
# throughout. Please refer to the TensorFlow dockerfiles documentation
# for more information.

This file has been truncated. show original

/cc @angerson