Welcome to Sovereign AI Hardware

Your data is about to come home.

Whether you are unboxing an EarthFrame One workstation, setting up an NVIDIA DGX Spark, or measuring your Mac with Warpt, this guide will help get your machine ready to run private, sovereign AI workloads in just a few minutes.

1 Warpt (Hardware & Power) 2 Mar (Fast Archives) 3 Machine Setup

Step 1: Check Your Hardware & Power with Warpt

Warpt is EarthFrame's lightweight tool for monitoring your computer's hardware and electricity usage. It runs completely locally on your system—no accounts, no setup, and no data ever sent to the cloud.

1A: Install Warpt

Install Warpt using Python's package manager (pip), or inside your preferred Python virtual environment:

# Install with pip
pip install warpt

# Or with uv
uv pip install warpt

1B: See Your Hardware & Live Power

Run a quick check to see what hardware is detected and how much power your machine is drawing:

# Show connected hardware, processors, and GPUs
warpt report

# Watch live power draw in watts (press Ctrl+C to stop)
warpt power

# Optional: save your machine's hardware summary to a file
warpt report --format json > machine_profile.json

Step 2: Install the Mar Archive Tool

Mar is a fast, modern file archive tool (like a next-generation zip or tar) built for large files and AI datasets.

Unlike traditional zip files that make you wait while unpacking gigabytes of data, Mar lets you look inside an archive and pull out individual files instantly, while automatically verifying that your files have not been corrupted.

2A: Download Mar for Your System

Choose your operating system below to copy the download commands. These make the mar command available in your terminal:

macOS • Apple Silicon (M1 / M2 / M3 / M4)

For modern Mac laptops and desktops:

# Download the Mac binary (v0.2.0)
curl -LO https://github.com/EarthFrame/mar/releases/download/v0.2.0/mar-macos-arm64
chmod +x mar-macos-arm64

# Move it so you can run 'mar' from anywhere
sudo cp mar-macos-arm64 /usr/local/bin/mar

# Test that it works
mar --help

Linux x86_64 • DGX Spark • EarthFrame One • Workstations

For standard Linux systems (Ubuntu, Debian, Fedora, Red Hat) on Intel or AMD processors:

# Download the Linux binary (v0.2.0)
curl -LO https://github.com/EarthFrame/mar/releases/download/v0.2.0/mar-linux-x86_64-musl
chmod +x mar-linux-x86_64-musl

# Move it so you can run 'mar' from anywhere
sudo cp mar-linux-x86_64-musl /usr/local/bin/mar

# Test that it works
mar --help

Build from Source • ARM64 Linux • Custom OS

For ARM64 Linux servers (such as NVIDIA Grace Hopper or Raspberry Pi) or building from source code:

# Clone the repository
git clone https://github.com/EarthFrame/mar.git
cd mar

# Install build dependencies:
# On Ubuntu/Debian: sudo apt install build-essential cmake libzstd-dev liblz4-dev libbz2-dev zlib1g-dev libdeflate-dev
# On Mac (Homebrew): brew install gcc zstd lz4 bzip2 libdeflate

# Compile and install
make
sudo cp mar /usr/local/bin/mar

# Test that it works
mar --help

2B: Try It Out (3 Quick Commands)

Here is how simple it is to package files, check what is inside, and unpack them:

# 1. Pack a folder into a single .mar file
mar create my_data.mar ./sample_folder/

# 2. View what's inside without having to unpack it
mar list my_data.mar

# 3. Unpack all files, or unpack just one specific file
mar extract my_data.mar
mar extract my_data.mar sample_folder/model.safetensors

Step 3: Machine-Specific Setup

Select your machine type below for a few quick checks to ensure your hardware, drivers, and power are ready:

Multi-GPU Systems • High-Density Compute

NVIDIA DGX / DGX Spark Setup

For multi-GPU machines, verify that your NVIDIA drivers, container system, and power monitoring are active:

1 Confirm GPUs and Drivers

Check that your NVIDIA graphics drivers and all installed GPUs are detected:

nvidia-smi

2 Test Docker GPU Access (Optional)

If you run models inside Docker containers, confirm Docker can talk to your GPUs:

docker run --rm --gpus all nvidia/cuda:12.4.1-base-ubuntu22.04 nvidia-smi

3 Watch Power and Temperature

Open a live monitor to watch real-time wattage and temperatures across all cards:

warpt power --interval 1 --watch

AI Workstation • Standard Wall Outlet (<1700W)

EarthFrame One Setup

EarthFrame One is engineered to deliver heavy compute while running safely on a single standard 120V / 15A wall circuit.

1 Wall Circuit Power Test

Run a 60-second power benchmark to verify that energy draw remains well within your wall outlet's capacity:

warpt power --benchmark --duration 60

2 Check GPU Connections

Confirm how your GPUs communicate with one another:

nvidia-smi topo -m

3 Check File Reading Speed

See how quickly data can be read directly from a Mar archive:

mar inspect --benchmark /path/to/archive.mar

Apple Silicon • M1 / M2 / M3 / M4

Mac Hardware & Power Observability

Warpt runs natively on Apple Silicon Macs, reading built-in hardware sensors without needing administrator (sudo) permissions.

1 See Your Mac's Hardware Overview

Inspect detected CPU cores, GPU cores, the Apple Neural Engine, and unified memory:

warpt report

2 Watch Live Mac Power Consumption

See live power usage in watts across CPU, GPU, and Apple Silicon chip rails:

warpt power

3 Test Mac GPU Acceleration (Optional)

If you use Python and PyTorch for AI, confirm that Apple's Metal GPU acceleration (MPS) is available:

python3 -c "import torch; print('MPS Available:', torch.backends.mps.is_available())"

Questions or Need a Hand?

Whether you have questions about setting up your hardware, power and outlet capacity, or organizing your data, the EarthFrame team is here to assist.

Contact EarthFrame Support →