The DGX Spark Homelab Setup Guide

About This Machine

The DGX Spark is not a rackmount server. It's a **tiny desktop unit** — 150 x 150 x 50.5 mm, 1.2 kg (2.6 lbs) — designed to sit on your desk next to your monitor. Think of it as a high-performance desktop that runs AI at datacenter scale.

Spec	Value
Form Factor	Small form factor desktop
Dimensions	150 x 150 x 50.5 mm
Weight	1.2 kg (2.6 lbs)
SoC	GB10 Grace Blackwell Superchip
CPU	20-core Arm (10 Cortex-X925 + 10 Cortex-A725)
GPU	Blackwell, 6144 CUDA cores, up to 1000 TOPS
Memory	128 GB LPDDR5x unified, 273 GB/s
Storage	1TB or 4TB NVMe M.2 (user-replaceable)
TDP	140W (SoC) / 240W total system
PSU	240W external (included)
Network	1x RJ-45 (10 GbE), ConnectX-7 Smart NIC, Wi-Fi 7
Ports	4x USB-C, 1x HDMI 2.1a
Expansion	None — no PCIe slots.

Power

The DGX Spark is incredibly power-efficient for the compute it delivers. Peak total system power is **240W** — that's less than a desktop gaming PC under load.

State	Power Draw	Notes
Idle	~50-80W	Barely anything — runs silently on your desk
Light inference	~100-150W	Single model, batch-size 1-4
Full GPU compute	150-240W	All Blackwell cores active, ARM cores at 100%
Peak	240W	Maximum — GB10 SoC 140W + system overhead 100W

💡 Note: nvidia-smi reports only GPU power, not total system power. The GB10 SoC TDP is 140W; the remaining ~100W powers the NIC, SSD, USB devices, and other components. Always use the included 240W PSU — lower power adapters will cause instability.

At 240W running 24/7 on average US electricity ($0.12/kWh): approximately $70/month at peak. Inference workloads average closer to $30-40/month. A standard 15A household circuit (1800W available) is more than sufficient.

Acoustics and Placement

The DGX Spark has **integrated thermal management** with no loud fans. It's designed for desk-side placement and operates almost silently — you'll notice it more by its compact footprint than by any sound it makes.

State	Acoustics	Placement
Idle	Almost inaudible	On your desk — no issues
Light inference	Quiet, barely noticeable	Desk or shelf
Full compute	Faint airflow hum — office-level	Any normal room

The machine is designed to operate at ambient temperatures of 5-30°C (41-86°F). Just ensure a few inches of clearance on all sides for airflow — that's it. No special mounting, no rack, no acoustic enclosure.

Physical Setup

Place it on your desk or shelf — that's literally where it belongs. No rack mounting required.
Plug in the 240W PSU and the power button.
Connect an HDMI display to the HDMI 2.1a port for initial setup.
Connect Ethernet to the RJ-45 (10 GbE) port for your homelab network.
Optionally add a second display via the HDMI port (the GB10 supports HDMI multichannel audio).

Storage

The DGX Spark ships with either a 1TB or 4TB NVMe M.2 drive with self-encryption. You can replace it yourself — just power off, remove the bottom cover, swap the M.2, and put it back. That's the only hardware access you get.

# Check your storage
df -h
lsblk

# If you want to add a second drive (USB or via M.2 slot replacement):
# The GB10 supports additional NVMe storage through the user-replaceable M.2 slot

Network Configuration

The DGX Spark has three network interfaces: a 10 GbE RJ-45 port, a ConnectX-7 Smart NIC (2x QSFP), and Wi-Fi 7. For most homelabs, the 10 GbE port is all you need.

Connecting to Your Homelab

# Check your network interfaces
ip link show

# Assign a static IP for homelab use
sudo ip addr add 192.168.0.100/24 dev eth0
sudo ip link set eth0 up

Connect the RJ-45 cable from the Spark to your homelab switch or router. That's it — no special switch required. The 10 GbE port works with any standard 10GbE or 2.5GbE switch (it auto-negotiates down).

The ConnectX-7 Smart NIC (2x QSFP) is for advanced multi-node clustering and GPU-direct networking. Most homelab users won't need it, but it's there if you want to build a Spark cluster later.

Firewall and Security

# Basic firewall
sudo ufw default deny incoming
sudo ufw default allow outgoing
sudo ufw allow ssh
sudo ufw allow 8080/tcp   # Ollama API
sudo ufw allow 443/tcp    # HTTPS
sudo ufw enable

# SSH key authentication
ssh-keygen -t ed25519 -C "spark-admin@gx10"
ssh-copy-id -i ~/.ssh/id_ed25519.pub spark@192.168.0.100

Docker / GPU Access

# Add NVIDIA repo
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | \
    sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
curl -s -L https://nvidia.github.io/libnvidia-container/$distribution/libnvidia-container.list | \
    sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
    sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

# Test GPU passthrough
docker run --rm --gpus all nvidia/cuda:12.2.0-base-ubuntu22.04 nvidia-smi

Monitoring

# Install node-exporter for system metrics
wget https://github.com/prometheus/node_exporter/releases/download/v1.7.0/node_exporter-1.7.0.linux-arm64.tar.gz
tar xzf node_exporter-*.tar.gz
sudo mv node_exporter-*/node_exporter /usr/local/bin/

# GPU metrics via DCGM exporter
docker run -d --name dcgm-exporter --gpus all -p 9400:9400 nvcr.io/nvidia/k8s/dcgm-exporter:3.3.6-3.4.0-ubuntu22.04

What Doesn't Work (And Why)

Common Idea	Why It Doesn't Work	Reality
Rack mount in a 42U cabinet	It's a desktop, not a 1U server	Just put it on your desk
PCIe expansion cards	No PCIe slots exist on the Spark	No expansion possible
Dedicated 20A circuit	Peak draw is only 240W	Any standard 15A circuit is fine
Liquid cooling	It has integrated thermal management, no liquid	Just needs a few inches of clearance
Sound-dampened enclosure	It's already almost silent	No acoustic treatment needed
PoE power	240W PSU required, not PoE	Plug in the included power supply

Summary Checklist

☐ Place on desk or shelf with clearance on all sides
☐ Connect the included 240W power supply
☐ Connect HDMI to a display for setup
☐ Connect Ethernet to your homelab network
☐ Set a static IP via Netplan
☐ Configure SSH keys
☐ Install NVIDIA Container Toolkit
☐ Verify GPU access with nvidia-smi

💡 Bottom line: The DGX Spark is a tiny desktop you plug in and go. The "setup" is literally: place it on your desk, plug in power, plug in Ethernet, plug in a display. Everything else is software. It's that simple.