tt-installer

Install the tenstorrent software stack with one command.

Quickstart

/bin/bash -c "$(curl -fsSL https://github.com/tenstorrent/tt-installer/releases/latest/download/install.sh)"

WARNING: Take care with this command! Always be careful running unknown code.

Using tt-metalium

In addition to our system-level tools, this script installs tt-metalium, Tenstorrent's framework for building and running AI models. Metalium is installed as a container using Podman (default) or Docker. You have two container options, both of which can be installed:

• tt-metalium container: 1GB, appropriate for using TT-NN
• tt-metalium Model Demos Container: 10GB, includes a full build of tt-metalium

Using the containers is easy- just run tt-metalium or tt-metalium-models. By default, this will launch the container with your home directory mounted so you can access your files. You can also run tt-metalium <command> to run commands inside the container, such as tt-metalium "python3".

For more about Metalium and TTNN, check out the examples page. For more information about the container, see this page on the wiki.

Using Python Tools

tt-installer installs two Python tools on your system:

1. tt-smi: Tenstorrent's System Management Interface
2. tt-flash: Utility to update your firmware

Running tt-smi launches the interface where you can see your hardware status and confirm the install worked properly.

Full List of Functions

tt-installer performs the following actions on your system:

1. Installs base packages the software stack depends on
2. Adds Tenstorrent package repositories to your package manager
3. Installs Tenstorrent software from the package repositories, including:
   • Kernel-Mode Driver (KMD)
   • System tools and HugePages configuration
   • Python packages (tt-flash, tt-smi, etc.)
4. Updates your card's firmware using tt-flash
5. Installs a container runtime (Podman by default, Docker optional)
6. Installs tt-metalium as a container and configures the wrapper script for convenient access
7. Installs tt-studio and tt-inference-server, our user-friendly model runtime systems

The installer will ask the user to make choices about Python environments and tt-metalium. If you wish to configure the installation more granularly, see Advanced Usage.

Advanced Usage

The installer supports command-line arguments for customization. For a full list of available arguments and their environment variable equivalents, please see this page on the wiki.

To install from a local file, download the latest install.sh:

curl -fsSL https://github.com/tenstorrent/tt-installer/releases/latest/download/install.sh -O
chmod +x install.sh
./install.sh

To see all available options:

./install.sh --help

To install without prompting the user:

./install.sh --mode-non-interactive

To skip certain components:

./install.sh --no-install-kmd --no-install-hugepages

To use Docker instead of Podman:

./install.sh --install-container-runtime=docker

To skip container runtime installation:

./install.sh --install-container-runtime=no

If you have already installed Docker or Podman, this option will leave them untouched.

To specify versions:

./install.sh --kmd-version=1.34 --fw-version=18.3.0

Note that the installer requires superuser (sudo) permisssions to install packages, add DKMS modules, and configure hugepages.

Distro Compatibility

Our preferred OS is Ubuntu 22.04.5 LTS (Jammy Jellyfish). Other operating systems will not be prioritized for support or features. For more information, please see this compatibility matrix:

OS	Version	Working?	Notes
Ubuntu	24.04.2 LTS	Yes	None
Ubuntu	22.04.5 LTS	Yes	None
Ubuntu	20.04.6 LTS	No	Deprecated
Debian	13.3	Yes	Curl is not installed by default
Fedora	43	Yes	May require restart after base package install
Fedora	42	Yes	May require restart after base package install
Other DEB-based distros	N/A	N/A	Unsupported but may work
Other RPM-based distros	N/A	N/A	Unsupported but may work
Arch Linux	N/A	No	Unsupported
NixOS	N/A	No	Unsupported, but some TT software is in nixpkgs (unofficial, use at own risk)