(Fanmade) Tenstorrent Stack Installer 🚀

Note: This is not an official script by Tenstorrent. This is a script that is fanmade, proceed at your own risk. Please visit the offical link for Tenstorrent's instructions for downloading/configuring software at https://docs.tenstorrent.com/getting-started/README.html

Automated Tenstorrent Full Stack Installation for Ubuntu 22.04

Inspired by the convenience of Lambda Stack, this repository provides a shell script (tenstorrent-stack.sh) designed to automate the complex setup process for the Tenstorrent software stack on Ubuntu 22.04 LTS.

The goal is to minimize the manual configuration and dependency wrestling required to get a Tenstorrent development environment up and running, covering system-level components, the TT-Metal/TT-NN SDK, and the TT-Forge SDK.

---

What it Installs

This script automates the installation and configuration of the following components:

1. System-Level Prerequisites:
   • Core development tools (build-essential, cmake, llvm, clang, etc. via the official TT-Metal install_dependencies.sh script).
   • dkms for kernel module management.
   • Rust (cargo).
2. Tenstorrent System Drivers & Tools:
   • Tenstorrent Kernel Module (TT-KMD): Installs the specified version via DKMS and loads the module.
   • Tenstorrent Firmware: Downloads the specified firmware bundle and uses tt-flash to attempt an update (requires reboot).
   • Tenstorrent Flash Utility (TT-Flash): Installs the utility via pip.
   • Tenstorrent Tools (tenstorrent-tools): Installs the .deb package to configure HugePages services (requires reboot).
   • Tenstorrent System Management Interface (TT-SMI): Installs the utility via pip for device monitoring.
3. Tenstorrent SDKs (Optional):
   • (Optional) TT-Metal / TT-NN SDK:
     • Clones the tt-metal repository (with submodules) from GitHub.
     • Builds the SDK from source (./build_metal.sh).
     • Creates a dedicated Python virtual environment (python_env) and installs Python dependencies.
   • (Optional) TT-Forge SDK:
     • Clones the tt-forge-fe repository from GitHub.
     • Note: Building TT-Forge requires separate manual steps after cloning (see Post-Installation).

---

Prerequisites

Before running the script, ensure you have:

• Operating System: Ubuntu 22.04 LTS (Jammy Jellyfish) is strongly recommended. The script includes a check but allows overriding for other Ubuntu versions at your own risk.
• Hardware: A system with compatible Tenstorrent hardware (e.g., Grayskull, Wormhole, Blackhole).
• Permissions: sudo / administrator privileges are required for system-level package installation, DKMS, firmware flashing, and service management.
• Internet Connection: Required to download packages, clone repositories, and fetch dependencies.
• Base Tools: git, wget, python3, pip3, cargo. The script checks for these.
• Disk Space: Significant free disk space is needed. Recommend ~30GB+ to accommodate repositories, build artifacts, and installed packages.

---

⚠️ Important Warnings & Considerations

• System Modifications: This script performs significant system-level changes, including installing kernel modules, potentially flashing device firmware, installing numerous packages via apt and pip (some globally with sudo pip3), and modifying system services (HugePages). Review the script carefully before execution. Use at your own risk.
• Multiple Reboots Required: The script will prompt you to reboot your system twice:
  1. After the firmware update attempt and TT-KMD installation.
  2. After configuring HugePages. These reboots are crucial for the changes to take effect properly.
• Version Compatibility: The script uses specific versions of TT-KMD, Firmware, and other tools defined in the configuration section. These versions were based on documentation available at a specific time. ALWAYS consult the official Tenstorrent SDK documentation and compatibility matrices for the versions appropriate for your specific hardware and SDK usage scenario. Using incompatible versions can lead to errors or unexpected behavior.
• sudo pip3 install: The script installs tt-flash, tt-smi, and (optionally) tt-topology system-wide using sudo pip3. If you prefer isolated environments, consider modifying the script to use pip3 install --user or manage these tools within virtual environments manually.
• Long Build Time: Building the TT-Metal SDK from source (./build_metal.sh) can take a very long time (potentially hours depending on your system).
• Not Idempotent: Running the script multiple times without cleaning up previous installations may lead to unexpected results or errors. It's designed for a fresh setup.

---

Installation

1. Clone the Repository:

   git clone https://github.com/cat768/tenstorrent-stack.git
   cd tenstorrent-stack

2. Make the Script Executable:

   chmod +x tenstorrent-stack.sh

3. Review the Script (Recommended): Open tenstorrent-stack.sh in a text editor and review the commands, versions, and installation paths.

4. Run the Script:

   ./tenstorrent-stack.sh

   The script will prompt for sudo password when needed and pause for the required reboots. Follow the on-screen instructions.

---

Post-Installation Steps

After the script completes successfully and you have performed the final reboot:

1. Verify System Components:

• Check KMD status: lsmod | grep tenstorrent (should show the module loaded)
• Check HugePages: cat /proc/meminfo | grep Huge
• Verify device detection: sudo tt-smi (should list your Tenstorrent devices)

2. Using TT-Metal / TT-NN (if installed):

• Navigate to the TT-Metal directory:

  cd ~/tenstorrent/tt-metal # Or your chosen INSTALL_DIR

• Activate the Python virtual environment:

  source python_env/bin/activate

• Set Required Environment Variables: (You might want to add these to your ~/.bashrc or ~/.profile for persistence within the activated venv context)

  export TT_METAL_HOME=$(pwd)
  export PYTHONPATH=$(pwd):$PYTHONPATH
  # Set ARCH_NAME based on your hardware: 'grayskull', 'wormhole_b0', or 'blackhole'
  export ARCH_NAME=<your_arch>

  Replace <your_arch> accordingly.
• Verify TT-NN installation (within activated venv):

  python3 -m ttnn.examples.usage.run_op_on_device

• (Optional) Set CPU governor for potentially better performance:

  sudo cpupower frequency-set -g performance

3. Using TT-Forge (if cloned):

• The script only clones the TT-Forge repository. You need to build it manually.
• Navigate to the TT-Forge directory:

  cd ~/tenstorrent/tt-forge-fe # Or your chosen INSTALL_DIR

• Follow the build instructions provided in the README.md or documentation within the tt-forge-fe repository. This typically involves installing specific prerequisites (like TVM) and using cmake and make.
• Alternatively, check the TT-Forge Releases page for pre-built Python wheel files (.whl). If available for your platform, you can install them using pip:

  # Assuming you downloaded the wheels after activating a suitable Python environment
  pip install <tvm_wheel_file>.whl <forge_wheel_file>.whl

• Verify installation (after building or installing wheels): Refer to the TT-Forge repository's testing instructions, e.g.:

  pytest forge/test/mlir/operators/eltwise_binary/test_eltwise_binary.py::test_add

---

Configuration

The script uses shell variables at the top to define versions and repository URLs (e.g., TT_KMD_VERSION, TT_FIRMWARE_VERSION, TT_METAL_REPO). Advanced users can modify these variables before running the script to target different versions, but ensure compatibility by checking official Tenstorrent documentation first.

---

Disclaimer

This script is provided "as is" without warranty of any kind. The author(s) are not responsible for any damage or data loss caused by its use. Use this script at your own risk. Always back up important data before performing significant system modifications.