Hello. This time I’ll note how to use TensorFlow 1.15 and TensorBoard with GPUs like RTX 3090 and 3080.

To use 3000 series GPUs with TensorFlow officially distributed, you need to upgrade to version 2. Since TensorFlow has API specification changes between version 1 and 2, running version 1 code is difficult.

So instead of using the officially distributed version, we’ll use the one distributed by NVIDIA.

pip install nvidia-pyindex
pip install nvidia-tensorflow
pip uninstall tensorboard
pip install nvidia-tensorboard

Requirements:

• Ubuntu 20.04 or later
• Nvidia Driver r455
• Python 3.8
• pip 19.0 or later

Now you can use TensorFlow==1.15 & TensorBoard even with RTX 3090!

References

Accelerating TensorFlow on NVIDIA A100 GPUs | NVIDIA Technical Blog

The NVIDIA A100, based on the NVIDIA Ampere GPU architecture, offers a suite of exciting new features: third-generation Tensor Cores, Multi-Instance GPU (MIG) and third-generation NVLink.

developer.nvidia.com

RTX3070などの30xx系GPU かつ CUDA11.3 インストール済みのホストPC上でTensorFlow v1.15.5を動かす

zenn.dev

GitHub - NVIDIA/tensorflow: An Open Source Machine Learning Framework for Everyone

An Open Source Machine Learning Framework for Everyone - GitHub - NVIDIA/tensorflow: An Open Source Machine Learning Framework for Everyone

github.com