This is a lightweight, real-time, browser-based monitoring dashboard for two NVIDIA DGX Spark nodes. Inspired by btop/htop and accessible from any browser. It is not for a single unit, if so, use btop instead.
It does not run directly on the DGX Sparks, instead run in another host monitoring two sparks over passwordless SSH connection.
It streams live metrics over WebSockets, pulls data via SSH from each node, and renders CPU, memory, GPU, disk, and network stats with sparklines and usage bars.
License: CC-BY-NC-ND-4.0
- 📡 Multi-node monitoring over SSH (no agents required)
- ⚡ Real-time updates via WebSockets (default: 1s refresh)
- 🧠 CPU: per-core usage + overall average + sparkline history
- 🧮 Memory: usage bar + historical sparkline
- 🎮 GPU: utilization, VRAM used/total, temperature (via
nvidia-smi) - 💽 Disks: capacity + live I/O (read/write rates) + sparkline
- 🌐 Network: per-interface throughput with autoscaled units
- 🔐 Key-based SSH only (no passwords stored)
+-------------------+ WebSocket +------------------------+
| Web Browser | <--------------------> | FastAPI Backend |
| (index.html, | | server.py |
| main.js, css) | | |
+-------------------+ | - AsyncSSH |
| - YAML config |
| - Metric collector |
+-----------+------------+
|
| SSH (keys)
v
+-----------------------+
| 2 DGX Spark Nodes |
| /proc, nvidia-smi, |
| sysfs, df, ip, etc. |
+-----------------------+
The simple backend does four main things:
- Loads
config.yamlto discover cluster nodes. - Opens SSH connections to each node using
asyncssh. - Runs a single batched shell command per node that collects:
- Load averages
- Memory usage
- CPU core stats (
/proc/stat) - GPU stats (
nvidia-smi) - Disk usage (
df) - Disk I/O (
/proc/diskstats) - Network counters (
/sys/class/net/*/statistics) - IP address of
enP7s7 - IP address
enp1s0f0np0, enp1s0f1np1(Hidden if not used)
- Streams JSON payloads to the browser over a WebSocket (
/ws) every second.
Runs one batched command per refresh to minimize SSH overhead and latency.
The frontend (index.html, style.css, main.js) has no build steps. Instead a single main.js for simplicity served by FastAPI:
- Opens a WebSocket to
/ws - Receives an array of node objects every second
- For each node:
- Renders usage bars (CPU, memory, GPU, disks, network)
- Maintains history buffers (up to 180 samples) for sparklines
- Draws sparklines on
<canvas>
- Automatically lays out nodes in two columns
- Backend connects to
ws://host:8000/ws - Backend loop:
- For each node:
- Connect over SSH
- Run batched collector command
- Parse sections: LOAD, MEM, CPU, GPU, DISK, NET, DISKIO
- Compute deltas for CPU, network, and disk I/O
- Send JSON payload to browser
- For each node:
- Frontend updates bars and sparklines
- 1 SSH round-trip per node per second (batched)
- Low backend CPU usage (text parsing + simple math)
- Small network footprint (aggregated metrics only)
- Lightweight canvas rendering in the browser
Scales well in my testing so far potentially extending to more nodes, primarily limited by SSH latency and network conditions.
- fastapi
- uvicorn
- asyncssh
- pyyaml
- websockets
There are minimal installation requirements. Perform the following steps on the monitoring hosts:
python3 -m venv .venv
source .venv/bin/activate
pip install -r backend/requirements.txt
On the monitoring host generate your public/private key pair. Example below for user 'sysadmin'.
ssh-keygen -t rsa -f ~/.ssh/id_rsa_DGX_dashboard
Generating public/private rsa key pair.
Enter passphrase (empty for no passphrase):
Enter same passphrase again:
Your identification has been saved in /home/sysadmin/.ssh/id_rsa_DGX_dashboard
Your public key has been saved in /home/sysadmin/.ssh/id_rsa_DGX_dashboard.pub
The key fingerprint is:
SHA256:blablabla/g sysadmin@monitoring.cisconian.net
The key's randomart image is:
+---[RSA 3072]---+
| || |
| || || || |
| || || || || || |
| || || || || || |
| || || || |
| || |
+-----[SHA256]----+
ssh-copy-id -i ~/.ssh/id_rsa_DGX_dashboard.pub sysadmin@<IP of DGX Spark node>
INFO: Source of key(s) to be installed: "/home/sysadmin/.ssh/id_rsa_DGX_dashboard.pub"
INFO: attempting to log in with the new key(s), to filter out any that are already installed
INFO: 1 key(s) remain to be installed -- if you are prompted now it is to install the new keys
Number of key(s) added: 1
Now try logging into the machine, with: "ssh 'sysadmin@'" and check to make sure that only the key(s) you wanted were added.
Let's make sure the SSH passwordless connection is working first!. From the monitoring host run:
ssh -i ~/.ssh/id_rsa_DGX_dashboard sysadmin@<IP of DGX Spark node>
Welcome to NVIDIA DGX Spark Version 7.4.0 (GNU/Linux 6.14.0-1015-nvidia aarch64)
System information as of Sat Feb 7 12:15:59 PM EST 2026
System load: 0.23 Temperature: 50.0 C
Usage of /: 29.1% of 3.67TB Processes: 513
Memory usage: 46% Users logged in: 0
Swap usage: 2% IPv4 address for enP7s7: 10.5.39.67
Last login: Sat Feb 7 12:14:15 2026 from monitoring.cisconian.net
It is working, so lets exit to go back.
exit
Create the configuration for each of the two nodes. Change the username, IP address and ssh key name accordingly.
nodes:
- name: Smarty67
host: 10.5.39.67
port: 22
username: sysadmin
key_path: /home/sysadmin/.ssh/id_rsa_DGX_dashboard
- name: Smarty69
host: 10.5.39.69
port: 22
username: sysadmin
key_path: /home/sysadmin/.ssh/id_rsa_DGX_dashboardServer runs on port 8000, but it can be changed. After activating the environment with:
source .venv/bin/activate
uvicorn server:app --host 0.0.0.0 --port 8000
INFO: Started server process [646972]
INFO: Waiting for application startup.
INFO: Application startup complete.
INFO: Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)
INFO: 10.5.37.24:62432 - "GET / HTTP/1.1" 200 OK
INFO: 10.5.37.24:62432 - "GET /favicon.ico HTTP/1.1" 200 OK
INFO: 10.5.37.24:58063 - "WebSocket /ws" [accepted]
INFO: connection open
INFO: 10.5.37.24:57319 - "GET /ws HTTP/1.1" 200 OK
INFO: connection closed
INFO: 10.5.37.24:57319 - "GET /favicon.ico HTTP/1.1" 200 OK
WebSocket client disconnected
INFO: 10.5.37.24:57319 - "GET /favicon.ico HTTP/1.1" 200 OK
INFO: 10.5.37.24:57857 - "WebSocket /ws" [accepted]
INFO: connection openOpen your browser and navigate to:
http://<server-ip>:8000
Real-time view of multiple NVIDIA DGX Spark nodes, showing CPU, memory, GPU, disk, and network activity with historical sparklines.
There are several improvements that can be done including code cleanup and scaling over two nodes primarily.
- More than 2 nodes on the same dashboard
- Multi-GPU support per node
- Per-process GPU usage (via
nvidia-smi pmonor compute apps query) - Threshold-based alerts (temp, memory, disk, CPU)
- Configurable refresh interval
- Historical time ranges (5 min, 1 hour, 24 hours)
- Optional metrics persistence (SQLite)
- Pluggable collectors (enable/disable modules)
- Collapsible sections
- Cluster-wide aggregate views
- Authentication and access control
- Integration with SIEM (Splunk, ELK, Graylog)
CC-BY-NC-ND-4.0
Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International