--- myst: html_meta: product-name: TT-QuietBox Blackhole™, Blackhole™ Networked AI Processor, Tensix core technology-concepts: PCIe, QSFP-DD, Installation, Setup, Electrostatic Discharge (ESD), Basic Input/Output System (BIOS) document-type: Task-Based Guide (How-To) --- ```{figure} ./qb2-setup-hero2.jpg :width: 65% :target: ./qb2-setup-hero2.jpg ``` # Hardware and Software Setup *Note: This product is pre-launch and documentation is subject to change.* This guide shows users how to safely unbox, setup hardware, and install software on a TT-QuietBox 2 (Blackhole) Workstation. ## Before You Begin Choose a suitable location for your System: * Choose a stable area for the TT-QuietBox 2 where it will not need to be moved regularly, and none of the vents will be blocked. * The Power Supply Unit (PSU) on the TT-QuietBox 2 is rated to draw up to 1600W of power. A standard 15A circuit can handle up to 1800W. When choosing the location of your TT-QuietBox 2, be mindful of the other electronics which may draw power from the same circuit. We recommend putting the TT-QuietBox 2 on a dedicated circuit, if not you may trip a breaker. * Review the {ref}`safety warnings `. * Inspect your package for signs of damage. Do not proceed with unboxing or installation if you suspect shipping damage to the system. Contact Tenstorrent support by [raising a support request.](https://tenstorrent.atlassian.net/servicedesk/customer/portal/1) Our team will review your request and provide assistance. ### Required Tools Ensure you have everything you need to get started. The Tenstorrent TT-QuietBox 2 (Blackhole) (TW-04003) package includes the following items: * 1x TT-QuietBox 2 (Blackhole) Workstation * 1x Power Supply Cord (C19 to NEMA 5-15P) * 1x AnkerWork S500 speakerphone For setup, you will also need your own: * Keyboard * Mouse * HDMI cable + monitor Note: Only use certified HDMI cables with the TT-QuietBox 2. Using non-certified HDMI cables may result in device damage, or non-compliant video operation. --- ## Step 1: Unbox the Workstation 1. Position the box in your prepared unboxing area, ensuring ample space to work around it. ```{figure} ./qb2-unboxing-box.jpg :width: 40% ``` 2. Remove the white side clips near the bottom of the packaging. Squeeze the tabs on the clip and pull to release. ```{figure} ./qb2-unboxing-clips.jpg :width: 40% ``` ```{figure} ./qb2-unboxing-clip-release.jpg :width: 40% ``` 3. Lift box lid, remove accessories box and styrofoam, then remove internal cardboard case and plastic sleeve. ```{figure} ./qb2-unboxing-lift-sleeve.jpg :width: 40% ``` ```{figure} ./qb2-unboxing-remove-accessories.jpg :width: 40% ``` ```{figure} ./qb2-unboxing-lift-inner-box.jpg :width: 40% ``` ```{figure} ./qb2-unboxing-lift-plastic-sleeve.jpg :width: 40% ``` 4. Lift the TT-QuietBox 2 out of the package using the handles. ```{figure} ./qb2-unboxing-lift-workstation.jpg :width: 40% ``` 5. Remove protective wrapping on all sides. ```{figure} ./qb2-remove-plastic-front.jpg :width: 40% ``` ```{figure} ./qb2-remove-plastic-side.jpg :width: 40% ``` --- ## Step 2: Set Up the Hardware ```{figure} ./qb2-rear-view.jpg :width: 60% ``` 1. **Connect the power cable.** Connect the provided C19 power cable to the workstation and then to a dedicated power outlet. See the Electrical Safety section for the full list of power requirements. 2. **Connect peripherals.** Connect the HDMI monitor, keyboard, and mouse to the back of the Workstation. (Please note: video is not supported through the USB-C port). For internet connections, we recommend Ethernet over WiFi for faster downloading of models. If you prefer Ethernet, connect your Ethernet cable to the RJ45 port. For sound and interaction with future text-to-speech (TTS) and speech-to-text (STT) AI models, there are a few options. You may either connect the provided speakerphone to the USB-C port or connect an audio device to the audiojack. 3. **Power on the Workstation.** On the back of the workstation, flip the switch on the PSU to the "I" position. 4. **On the front of the Workstation, press the power button to turn the system on.** ```{figure} ./qb2-power-button.jpg :width: 40% ``` --- ## Step 3. First-Time Log In Once your system is booted up, the login prompt will appear. Enter the default password: **ttuser** ## Step 4. Change Default Password After logging in, open a Terminal window by pressing Ctrl+Alt+T and follow these steps to change your password from the public default: 1. Run command: `passwd` 2. Enter current password: **ttuser** 3. Enter new password of your choosing and hit enter to confirm the change. ## Step 5. Setup and Confirm Internet Connection TT-QuietBox 2 can be connected to the internet via WiFi or Ethernet cable. For faster model downloads, we recommend a direct Ethernet connection. If you have not done so already during hardware setup, connect your Ethernet cable to the standard RJ45 port on the back of the Workstation. Then, verify your internet connection by clicking on the status icons in the upper right corner of the screen. Confirm the internet connection reads "Wired." If you would prefer to set up a WiFi connection, on your monitor, click on the status icons in the top right corner of the screen. Then, click on "Wi-Fi" (it may say "not connected" or "off"). Select your WiFi network from the drop-down list, enter the password, and click "Connect." ## Step 6. Update Ubuntu Operating System TT-QuietBox 2 comes pre-installed with the Ubuntu operating system (24.04.3 LTS). **Step 1:** Upon logging in, a Ubuntu Software Updater may offer a prompt that new software has been issued since the latest release. If this prompt appears, click "Install Now" to download the latest Ubuntu updates. **Step 2:** To ensure you have the latest system package updates, open a Terminal window by pressing Ctrl+Alt+T and run: ```bash sudo apt update && sudo apt upgrade -y ``` Wait for any downloads to complete, then proceed to the next step. --- ## Step 7: Verify System Recognition of Blackhole Cards To verify all cards are up and running in your TT-QuietBox, launch TT-SMI. This is Tenstorrent's simple command line utility that displays devices, device telemetry and other system information. 1. In the terminal your shell prompt should read: `(.tenstorrent-venv) ttuser@tt-quietbox:~$` 2. Run command `tt-smi` 3. Under the “Device Information” pane, you should see an output which lists four recognized accelerators. See the screenshot below for reference. ```{figure} ./screencap-tt-smi-qb2.jpg :width: 80% ``` :::{admonition} Important :class: caution If `.tenstorrent-venv` is not active, or you don’t see all four accelerators listed in the Device Information pane of TT-SMI, please [visit the troubleshooting page.](./support-bh-2.md) ::: 4. Once all cards have been verified, close TT-SMI by pressing Q on your keyboard. --- ## Step 8: Get Access to Model Weights TT-QuietBox 2 comes pre-installed with TT-Studio, Tenstorrent's simple web interface for running AI models. TT-Studio currently supports the following common models on TT-QuietBox 2: | Type | Model | | --- | --- | | Video Gen | Wan 2.2 (coming soon) | | Text to Image | Flux (coming soon) | | Language Models | GPT-OSS 120B (coming soon), Llama 3.1 70B, Qwen3-32B, Llama 3.1 8B | To deploy a model in TT-Studio, first download the model weights from Hugging Face. Note: for Qwen3-32B, the model weights come pre-downloaded onto your TT-QuietBox 2. However you will still require an access token from Hugging Face to use the model. Hugging Face is a free, open source community for collaborating on AI models and applications. Hugging Face access tokens are the unique security keys that allow weights from AI models to be downloaded to your machine. Read more about how user access tokens work in the [Hugging Face documentation](https://huggingface.co/docs/hub/en/security-tokens#how-to-manage-user-access-tokens). To get access to model weights on Hugging Face, follow these steps: 1. Open a new browser window and navigate to [huggingface.co](https://huggingface.co). 2. Create or log in to your Hugging Face account. 3. Create your access token at [huggingface.co/settings/tokens](https://huggingface.co/settings/tokens). Copy the access token as it is only displayed once and will be needed in the next step. 4. On the Hugging Face website, visit the model page of your choice. Depending on your choice of model you may need to click **Request Access** in the upper right corner or you may be prompted to scroll through and sign a community license agreement. ## Step 9: Launch TT-Studio 1. Ensure you have the latest version of TT-Studio from the open source Tenstorrent GitHub repository. To do this, open a Terminal window by pressing Ctrl+Alt+T and run the following command: ```bash cd ~/.local/lib/tt-studio git pull ``` 2. When the process is complete, run this command in Terminal to open TT-Studio: ```bash tt-studio ``` Your Terminal window will show a "Welcome to TT-Studio" message. See screenshot below for reference. ```{figure} ./qb2-screenshot-open-tt-studio.jpg :width: 80% ``` ## Step 10: Launch Your First Model At the bottom of the Terminal screen, you will be prompted to enter a Hugging Face User Access Token (aka "HF_TOKEN"). 1. Paste this Hugging Face token into the Terminal window and run the command. 2. If the system asks, “Do you want to install dependencies using Docker?” enter “Y” for “yes.” Installing dependencies may take about 3 minutes. ```{figure} ./qb2-screenshot-install-docker.jpg :width: 80% ``` 3. When prompted, enter your sudo password (this is the same password you use to log in to your Workstation). TT-Studio runs on top of TT-Inference Server which requires sudo privileges to set up. ```{figure} ./qb2-screenshot-sudo-pw.jpg :width: 80% ``` 4. The TT-Studio web app will now launch in your default web browser. Click on the model of your choice from the drop-down menu and press "NEXT." ```{figure} ./qb2-screenshot-select-model2.png :width: 80% ``` 5. When prompted on the next screen, proceed by hitting "DEPLOY." The model weights will start downloading automatically. Downloading a model can take anywhere from a few minutes to a few hours, depending on the model you’ve selected and the speed of your internet connection. WiFi connections will be slower than direct Ethernet. TT-Studio will show the status of "model unavailable" until model download is finished. ## Other Methods of Running Models After your TT-QuietBox 2 is set up, feel free to explore other methods of running models on other layers of Tenstorrent's software stack. You may want to: - Run models directly from Terminal using [TT-Inference-Server](https://github.com/tenstorrent/tt-inference-server), the fastest way to deploy and test models for serving inference on Tenstorrent hardware. - Run [model demos](https://docs.tenstorrent.com/getting-started/model-demos.html) using TT-Metalium. --- ## Need Additional Support? If you encounter any issues, or have a question that isn't covered in this guide, please [raise a support request.](https://tenstorrent.atlassian.net/servicedesk/customer/portal/1) Our team will review your request and provide assistance.