Python3 Machine Learning Workspace for NVIDIA RTX 5000 series GPUs

This repository contains configuration for a GPU-accelerated machine learning environment using Docker and Jupyter Lab for Windows 11, specifically optimized for NVIDIA RTX 5000 series GPUs for AI training and inference.

Python version: 3.12.3

---

Development System Specs

This workspace was developed and tested on:

• OS: Microsoft Windows 11 Home
• CPU: Intel® Core™ i5-13450HX
• GPU: NVIDIA® GeForce RTX 5060
• Storage: 512 GB SSD
• Memory: 16 GB RAM
• WSL: Ubuntu 24.04

---

What's Included

• Dockerfile: Defines the container image with TensorFlow + CUDA support
• docker-compose.yml: Defines the service for local development
• rtx-5060_dev-requirements.txt: Python dependencies for ML workflows

---

Why WSL2 Ubuntu + NVIDIA TensorFlow CUDA Image

TensorFlow no longer supports native GPU acceleration on Windows after version 2.10, so you cannot reliably use CUDA/TensorFlow GPU features directly on Windows.

To train models with GPU support (e.g., on an RTX 5060), you need a Linux environment such as WSL2 Ubuntu, where Docker + NVIDIA drivers can expose the GPU to containers.

For this reason:

• The base image nvcr.io/nvidia/tensorflow:25.02-tf2-py3 is an official NVIDIA container image that includes pre-installed TensorFlow + CUDA + cuDNN and all the correct GPU libraries
• This ensures optimal compatibility and saves you from manually configuring CUDA, cuDNN, and TensorFlow

---

Prerequisites

Before running this workspace, ensure you have completed the following:

Required Setup

• NVIDIA Driver (latest version for RTX 5000 series)
• WSL2 with Ubuntu 24.04 installed
• Latest Docker Desktop
• CUDA Toolkit installed on WSL2

Verification Steps

1. Verify NVIDIA driver and CUDA are working in WSL2:

   nvidia-smi

   This should display your GPU information and CUDA version.

2. Verify CUDA Toolkit installation:

   nvcc --version

   This should display the CUDA compiler version.

Setup Tutorial

If you haven't set up the above requirements, follow this comprehensive tutorial:

WSL2 + NVIDIA Driver + CUDA Toolkit

This video covers:

• Installing and updating NVIDIA drivers
• Setting up WSL2 Ubuntu
• Installing CUDA Toolkit in WSL2
• Verifying GPU access in Docker

---

Build and Run

1. Build the container (make sure you're in subdirectory: dockerfiles)

docker compose build

2. Start the service

docker compose up -d

3. Access Jupyter Lab

Open your browser at:

http://127.0.0.1:8888/lab?token=

Token: rtx-5060_dev

4. Shutdown the container

To stop and remove the container:

docker compose down

---

Project Structure

python3-workspace/
├── dockerfiles/
│   ├── Dockerfile
│   ├── rtx-5060_dev-requirements.txt
│   └── docker-compose.yml
└── README.md

File Descriptions

Dockerfile:

• Uses NVIDIA TensorFlow + CUDA base image (nvcr.io/nvidia/tensorflow:25.02-tf2-py3)
• Installs system dependencies (python3-dev, git, graphviz, OpenGL libraries)
• Copies and installs Python ML dependencies from rtx-5060_dev-requirements.txt
• Sets TensorFlow optimization environment variables
• Exposes port 8888 for Jupyter Lab

docker-compose.yml:

• Defines the tf-lab service with GPU access enabled
• Builds from the Dockerfile in the dockerfiles/ directory
• Maps the parent directory (python3-workspace/) to /workspace inside the container
• Configures Jupyter Lab to run on 0.0.0.0:8888 with token authentication
• Creates a persistent volume for Jupyter settings
• Sets resource limits for optimal GPU performance (IPC host mode, memory lock)

rtx-5060_dev-requirements.txt:

• Contains Python package dependencies for ML workflows:
  • numpy<2 - Numerical computing (version constraint for compatibility)
  • torch, torchvision, torchaudio - PyTorch deep learning framework
  • pandas - Data manipulation and analysis
  • scikit-learn - Machine learning algorithms
  • matplotlib, seaborn - Data visualization
  • opencv-python - Computer vision library
  • pydot - Graph visualization
  • jupyterlab, ipykernel - Interactive notebook environment

---

Important Notes

• This Docker container is dedicated for RTX 5000 series GPUs for AI training and inference in WSL2
• TensorFlow GPU support inside Docker only requires the NVIDIA driver on the host; the CUDA Toolkit itself is provided inside the container image
• NVIDIA's container images from catalog.ngc.nvidia.com/orgs/nvidia/containers/tensorflow include CUDA, cuDNN, and TensorFlow correctly configured to work together, reducing setup and compatibility issues
• The container exposes Jupyter Lab on port 8888
• Python 3.12.3 is used inside the container
• Always ensure nvidia-smi works properly in WSL2 before running the Docker container