Having trouble connecting to your NVIDIA DGX Spark?

Follow these steps to get access remote and start exploring

If you tried to set up your Nvidia DGX Spark remotely and you're now unable to SSH into the device, there is a chance I have a solution for you.

If for whatever reason during the install your spark SHH service experienced an error, like it did in my case. You will not be able to connect to the device until you connected to a monitor, keyboard and mouse, and login into the device.

The simplest way that I found was to go to Best Buy and get a $10 keyboard and a $10 mouse with a USB-C dongle and then connect that to Spark. If you don't have a compatible screen, use any basic TV with an HDMI cable. If you have to, BestBuy sells monitors for like $200; worth a visit.

Once you have connected these peripherals and you got into the Spark, here's what she do:

1. Regenerate the Missing Host Keys

Run this command to automatically generate all standard SSH host keys (RSA, ECDSA, Ed25519, etc.) in `/etc/ssh/`:

sudo ssh-keygen -A

This is safe and non-destructive; it only creates keys that don’t exist. If prompted, accept the defaults (no passphrase needed for host keys).

2. Fix Permissions on SSH Files (If Needed)

Ensure the keys and config have correct ownership and permissions (owned by root, readable but not writable by others):

sudo chown -R root:root /etc/ssh
sudo chmod 600 /etc/ssh/ssh_host_*_key
sudo chmod 644 /etc/ssh/ssh_host_*_key.pub
sudo chmod 644 /etc/ssh/sshd_config

3. Restart the SSH Service

This will just make sure the service starts with new configurations

   sudo systemctl restart ssh

4. Verify the Fix

Check the service status (should now show “active (running)” without errors)

 sudo systemctl status ssh

Confirm the port is listening:

sudo ss -tuln | grep :22

After this, my Spark became instantly available on the network and I was able to use NVIDIA Sync application to connect with it. You can also ssh directly into the spark.

ssh your_spark_username@your_spark_local_network_ip

That’s it, it’s that simple. Unfortunately it was not as simple without a direct connection to the Spark system to see what was going on inside.

Here’s me explaining why the initial Spark setup is rather user-unfliendly:

— Kirill

p.s. If you like this type of content, please join me on X Nvidia Spark Group