Enabling NVIDIA® GPUs for accelerating deep learning model training

Prerequisites

Before using NVIDIA® GPUs, ensure that you meet the following prerequisites:

System requirements
- Ubuntu 22.04 LTS
Hardware requirements
- NVIDIA® GPUs that meet the specifications outlined by TensorFlow in this link.

Step-by-step instructions

Option 1 (Recommended)

Step 1. Install NVIDIA® GPU drivers

Install the NVIDIA® GPU drivers with a minimum version of 525.60.13.

For example, you can download and install the driver of 535.274.02 with the following commands.
```
wget https://download.nvidia.com/XFree86/Linux-x86_64/535.274.02/NVIDIA-Linux-x86_64-535.274.02.run

sudo sh NVIDIA-Linux-x86_64-535.274.02.run
```
Next, verify the installation using:
```
nvidia-smi
```
Step 2. Install CUDA Toolkit 12.5

Time Series Studio requires the CUDA Toolkit 12.5. You can install and config the CUDA Toolkit as follows.
```
wget https://developer.download.nvidia.com/compute/cuda/12.5.0/local_installers/cuda_12.5.0_555.42.02_linux.run
sudo sh cuda_12.5.0_555.42.02_linux.run
```
Next, you need to add the following lines to your ~/.bashrc file.
```
export PATH=/usr/local/cuda-12.5/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-12.5/lib64:$LD_LIBRARY_PATH
```
After you have added them to the ~/.bashrc file, execute source ~/.bashrc and sudo ldconfig.

Step 3. Install cuDNN 9.3

Time Series Studio also requires cuDNN 9.3. You can install the cuDNN as follows.

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt-get update
sudo apt-get -y install cudnn
sudo apt-get -y install cudnn-cuda-12

After installing the CUDA Toolkit and cuDNN, you can leverage the GPU to train integrated deep learning models in eIQ Time Series Studio. Please note that if either component is installed incorrectly, for example, if the version is incompatible, the GPU will not function properly during training.

Option 2

Step 1. Install NVIDIA® GPU drivers

Install the NVIDIA® GPU drivers with a minimum version of 525.60.13.

For example, you can download and install the driver of 535.274.02 with the following commands.
```
wget https://download.nvidia.com/XFree86/Linux-x86_64/535.274.02/NVIDIA-Linux-x86_64-535.274.02.run

sudo sh NVIDIA-Linux-x86_64-535.274.02.run
```
Next, verify the installation using:
```
nvidia-smi
```
Step 2. Install Python 3.10.8

If you already have Python 3.10.8 installed, you can skip this step.

If Python 3.10.8 is not installed, use a Python version management tool such as pyenv.

Example:
```
pyenv install 3.10.8
# set Python 3.10.8 globally
pyenv global 3.10.8
```
Step 3. Create a virtual environment with venv

Navigate to your preferred directory and create a virtual environment named tss using Python 3.10.8:
```
python -m venv tss
```
Activate the environment:
```
source tss/bin/activate
```
Step 4. Install TensorFlow

Upgrade pip to the latest version:
```
pip install --upgrade pip
```
Then, install TensorFlow with CUDA support:
```
pip install tensorflow[and-cuda]==2.18.1
```

Step 5. Configure environment variables

After installing TensorFlow, locate the nvidia folder under your virtual environment (e.g., lib/python3.10/site-packages/nvidia).

Add the following lines to your ~/.bashrc file, replacing /your_path/ with the actual path:

export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:$(find /your_path/tss/lib/python3.10/site-packages/nvidia -type f -name "*.so*" -exec dirname {} \; | sort -u | paste -sd: -)"

export XLA_FLAGS="--xla_gpu_cuda_data_dir=/your_path/tss/lib/python3.10/site-packages/nvidia/cuda_nvcc"

Reload the configuration:

source ~/.bashrc