Here's a step-by-step guide to setting up an On-Prem open source GPT server with dual boot capabilities, featuring Windows 11 and Ubuntu. This article focuses on installing necessary software and ML libraries, such as CUDA.
Windows 11 Installation and Setup
-
CUDA Installation:
- Confirm your GPU is CUDA-compatible:
- Open a run window from the Start Menu.
- Execute:
control /name Microsoft.DeviceManager
- Refer to NVIDIA's CUDA GPUs list to check if your GPU is listed.
- Install CUDA from NVIDIA's installation guide.
- Download the CUDA Toolkit.
- Install required C++ libraries via Microsoft Visual Studio Community.
- Verify your CUDA installation by building and running CUDA samples. Access them from NVIDIA's GitHub.
-
System Configurations:
- Rename your PC in
System settings > Rename this PC.
- For domain integration, refer to
System properties > Domain or Workgroup.
-
Python and Libraries:
- Install Python (Run as administrator).
- For git bash compatibility with Python, modify
.bashrc to include the alias: alias python="winpty python.exe"
- Install ML libraries:
- TensorFlow:
pip install tensorflow
- PyTorch: Follow guidelines on pytorch.org for your environment.
- For Hugging Face Transformers library, refer to the official installation guide.
-
Security and Disk Management:
- Regularly run Windows updates.
- Disable BitLocker for dual boot:
- Navigate to Control Panel > System and Security > Bitlocker Drive Encryption.
- Save the encryption key, then decrypt the C: drive.
- Ensure the D: drive is empty for Ubuntu installation.
-
User Management and Network Configuration:
- Create a local user via
Accounts > Other Users.
- Enable Remote Desktop and configure firewall settings for RDP.
Ubuntu Installation and Setup
-
Ubuntu Installation:
- Boot from a USB with Ubuntu. Press
F9 to select the boot order.
- Adjust startup options for display compatibility by adding
nomodeset in the GRUB menu.
- Proceed with Ubuntu 22.04.3 installation. Ensure you select the correct partition to avoid data loss.
-
CUDA Installation:
- Follow NVIDIA's guide for Linux.
- Ensure your GPU is CUDA-compatible and that the OS is supported by the CUDA Toolkit.
- Install the required gcc version and Linux kernel headers.
- Download and install the CUDA toolkit suitable for your environment.
-
Python and Libraries:
- Install pip for Python.
- Follow the same steps under the Windows section to install TensorFlow and PyTorch.
-
System Configurations:
- Set the hostname with
hostnamectl set-hostname [computername].cuit.columbia.edu.
- Run software updates regularly.
- Set Ubuntu as the default OS on startup.
-
User Management and Network Configuration:
- Create a user account.
- Enable the firewall with
sudo ufw enable.
- Install and configure SSH for security:
- Rate limit connections with
sudo ufw limit ssh.
- Modify
sshd_config to disable port forwarding and allow specific users.
- Optionally, use
nmap to scan for open ports and security vulnerabilities.
HP Z8 Fury G5 Workstation Technical Specifications:
- CPU: Intel Xeon W-3200 series
- GPU: NVIDIA Quadro RTX 6000
- RAM: 128GB DDR4 ECC]
- Storage:
- Primary: 1TB NVMe PCIe M.2 SSD
- Secondary: 2TB 7200 RPM SATA HDD
- Additional Features: "Dual 10GbE LAN ports", "Thunderbolt 3 ports"
Note: The above specifications are placeholders and need to be replaced with the exact details of the particular HP Z8 Fury G5 workstation in use.
You can typically find the exact specifications for your HP Z8 Fury G5 workstation in its user manual or on its product page on HP's official website.
References:
Ensure you always refer to official documentation and seek expert advice when uncertain about a configuration or setting.
Recognition: A big thank you to Osman Kabir - Senior Applications Systems Admin CUIT, David Chan - Senior Applications Systems Developer CUIT, & Basem Aly - Assistant Director, Instructional Technology Columbia Law School for their huge contribution in helping to create and document the On-Prem open source GPT server.
