Hey NVIDIA Developer Community,
As someone running DGX Spark for quant/AI workflows, I’ve completely transformed how I interact with my system. No longer chained to the desk—DGX Dashboard on iPhone, JupyterLab on iPad, and Termius terminal for tailing/screen reattach—all powered by Termius port forwarding and Tailscale for seamless anywhere access.
This setup gives incredible flexibility:
-
Quick resource checks on iPhone while away.
-
Full notebook prototyping on iPad from anywhere.
-
Terminal commands for screen -r or tail -f spark.log on the go.
-
Even at your desk: Keep DGX Dashboard open on iPhone next to your monitor for instant metrics without alt-tabbing—super convenient for monitoring during long runs.
Why Termius Stands Out (Disclosure: No Affiliation)
(Full disclosure – I have no connection or affiliation with Termius; I’m just a very satisfied user.)
Termius has become my go-to for a few key reasons:
-
Excellent port forwarding UI – rules are easy to create and manage (e.g., Dashboard 11000, JupyterLab 11003).
-
Reliable SSH + Mosh – sessions stay alive over spotty cellular or Wi-Fi switches.
-
Clean, modern interface – fast SFTP browser, snippets, AI command suggestions (Pro), and great external keyboard support.
-
Native apps across platforms – Full-featured desktop apps for macOS, Windows, plus dedicated iOS/iPadOS apps (universal for iPhone and iPad) and Android.
-
Cross-platform sync – hosts and rules sync between iPhone, iPad, and desktop apps.
That said, it’s not the only option. Other strong SSH clients worth trying:
-
Secure ShellFish – outstanding Files app integration for mounting DGX filesystem natively (double-tap files to edit).
-
Blink Shell – Mosh-focused, great for developers who want a more code-centric experience.
-
Shelly or Prompt – lighter alternatives if you need basic SSH without extras.
Termius just hits the sweet spot for me with reliability and polish on iOS/iPadOS.
Core Setup (Simple & Secure)
-
Termius (iOS/iPadOS):
-
Add DGX host (local IP or Tailscale IP).
-
Port forwarding rules:
-
Dashboard: Local 11000 → localhost:11000
-
JupyterLab: Local 11003 → localhost:11003
-
-
Open Safari → localhost:11000 (Dashboard) or localhost:11003 (JupyterLab).
-
-
Tailscale (official NVIDIA recommendation – see https://build.nvidia.com/spark/tailscale):
-
Install on DGX and mobile devices → join same tailnet.
-
Zero-config, encrypted access from cellular, hotel Wi-Fi, anywhere.
-
No open ports, no complex VPN.
-
Real Flexibility Highlights
-
On the move: Tail a Spark job log via Termius terminal on iPhone during commute, reattach hung screen sessions from airport.
-
At desk convenience: iPhone propped up showing live Dashboard metrics (memory/GPU) while you code on main monitor—glance without switching windows.
-
iPad power: Magic Keyboard + JupyterLab = laptop-like prototyping anywhere, perfect for reviewing experiments on couch or travel
Termius terminal on iPhone – ready for tail/screen commands
This combo (Termius + Tailscale) is rock-solid and officially supported Tailscale solution. It’s liberated my workflow—no more rushing back for job checks.
Anyone else running mobile DGX access? What SSH client do you prefer, or any Tailscale tips for DGX Spark?
Cheers,
Mark
(DGX Spark • Quant/AI)