Starting Guide

Welcome to Tenstorrent! This guide will walk you through setting up your Tensix Processor(s) and installing necessary software.

Table of Contents

  1. Prerequisites

  2. Unboxing and Hardware Setup

  3. Software Installation

  4. Support & FAQ

1. Prerequisites

Before you begin, ensure you have the following:

  • A compatible host machine with minimum hardware and OS requirements as specified by each product’s minimum system requirements:

  • Network access to download software packages.

  • Administrator privileges on the host machine.

NOTE: The recommended OS for all Tenstorrent software is Ubuntu 22.04 LTS (Jammy Jellyfish). Each SDK may support newer distributions of Ubuntu; however, compatibility should be considered experimental at this time.

NOTE: Software support for Grayskull has been discontinued. The last supported versions of Tenstorrent’s software for Grayskull are as follows:

  • TT-Firmware: fw_pack-80.15.0.0.fwbundle

  • TT-KMD: ttkmd_1.31

  • TT-Buda: v0.19.3

  • TT-Metalium: v0.55

TT-QuietBox BIOS Requirement

The BIOS for the host motherboard is configured at the factory with the setting for PCIe AER Reporting Mechanism set to OS First. Tenstorrent’s TT-SMI software will fail if this setting is not configured properly. You should not have to change this setting when first setting up your TT-QuietBox.

If for whatever reason the BIOS needs to be updated or is reset, this setting must be configured again to ensure TT-SMI is able to function. It is located in the BIOS here:

Chipset -> AMD CBS -> NBIO Common Options -> NBIO RAS Common Options -> PCIe AER Reporting Mechanism

2. Unboxing and Hardware Setup

  1. Unpack the hardware and check all components against the provided list.

  2. Install the hardware following the hardware installation manual and safety guidelines by product:

  3. Secure the hardware in place, ensuring it is firmly seated and all connections are stable.

3. Software Installation

To interact with the Tensix Processor(s), you’ll need to install the system-level dependencies on your host machine.

Tenstorrent provides a bash script, tt-installer, for fast and easy setup of our software stack. The installer supports Ubuntu, Fedora, and Debian. To use it, paste the following into your terminal:

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

tt-installer configures necessary packages on your system and installs system-level tools as well as our development framework, TT-Metalium. By default, TT-Metalium is installed as a container using Podman. This containerized environment is appropriate for most users as explained here, but advanced users and developers may wish to install Metalium natively on the host system or use Docker instead of Podman. See TT-NN / TT-Metalium Installation for manual installation instructions.

For more information about tt-installer, please see the repository. If you would prefer to install the software stack manually, see Manual Installation.

Next Steps

After tt-installer finishes successfully and you have restarted your system, you can proceed to next steps. You may want to:

  • Run existing models like Llama and DeepSeek: Use TT-Transformers.

  • Use a high-level interface to build your own models or migrate from Torch: Use TT-NN.

  • Install TT-Metalium and write high-performance C++ kernels: Read First 5 Things below.

  • Learn more about our unique architecture: Start by reading this guide.

First 5 Things To Do With TT-Metalium

  1. Install and Build: If you’re a developer, install and build the project by following the instructions in the installation guide.

  2. Beginner TT-Metalium Usage | DRAM Loopback: Try creating a basic kernel example that uses the L1 and DRAM memory structures of the Tenstorrent device.

  3. Beginner TT-Metalium Usage | Eltwise Binary Kernel: Augment your loopback example with an additional kernel that will use the compute engine of the Tensix core to add values in two buffers.

  4. Beginner TT-Metalium Usage | Single-Core Matrix Multiplication Kernel: Use TT-Metalium to define your own matrix multiplication kernels. Refer to our simpler single-core example as a starting point.

  5. Advanced Metalium Usage | Multi-core Matrix Multiplication Kernel: Explore expert-level usage by building on the previous example to create a multi-core implementation.

Starting with TT-NN

To get started with TT-NN, visit this page.

TT-Buda (Deprecated)

Tenstorrent has discontinued development and support for the TT-Buda stack; these links are provided for reference and for developers still using Grayskull® cards.

Support & FAQ

For support, forums, and community, visit Tenstorrent’s Discord channel.

For additional support, file any issues through our Customer Success Platform or you can contact us directly at support@tenstorrent.com.