Hey Basu :)

You've got your own slice of Shivam's NVIDIA DGX Spark. Here's how to get in.

Get connected

Install Tailscale and sign in Download it from tailscale.com/download. A free account is fine.
Accept Shivam's invite Open this invite link, sign in with your Tailscale account, and click Accept:
https://login.tailscale.com/admin/invite/8ckPYZU6ZN6KQEgj9KdD11
The link works once, and only for you. The machine will show up in your Tailscale app as spark-2944-dgx-userspace.
Log in over SSH Open a terminal (PowerShell on Windows) and run:
ssh basu@spark-2944-dgx-userspace.tailc87c5a.ts.net

The first time, SSH asks "Are you sure you want to continue connecting?". Type yes. Then enter the password Shivam sent you separately. Nothing shows on screen while you type it, which is normal.
Set your own password (optional) Once you're in, run this and follow the prompts:
passwd

The address only works while Tailscale is connected on your machine. If the name doesn't resolve, use the IP instead: ssh basu@100.95.182.62

CPU	10 coresCores 10-19: 5 fast + 5 efficient.
Memory	60 GB, RAM + GPU memory combinedThe Spark's CPU and GPU share the same memory. If you go over 60 GB, your largest process is stopped.
Storage	1.8 TB in /home/basuReserved just for you. Keep your files here.
Temp files	20 GBCovers /tmp and /var/tmp, for temporary files only.
GPU	NVIDIA GB10, shared with ShivamJobs running at the same time share the GPU's compute.

You're a regular user, so there's no sudo or Docker. Install everything inside your home folder. On this machine, pip install --user is blocked by Ubuntu, so use a virtual environment:
python3 -m venv ~/venv && source ~/venv/bin/activate
Or install uv, which doesn't need sudo:
curl -LsSf https://astral.sh/uv/install.sh | sh
CUDA is already installed and nvcc is on your PATH. The machine is ARM64 (aarch64), so get the ARM64 + CUDA build of any GPU packages, such as PyTorch.

Run these from your own computer, not the Spark. To copy a file to the Spark:
scp myfile.zip basu@spark-2944-dgx-userspace.tailc87c5a.ts.net:~/
To copy a file back to your computer:
scp basu@spark-2944-dgx-userspace.tailc87c5a.ts.net:~/results.csv .
For whole folders, use scp -r or rsync -av. VS Code Remote - SSH also lets you drag and drop files.

If you don't have an SSH key yet, make one by running this and pressing Enter at every prompt:
ssh-keygen -t ed25519
On Mac or Linux, send the key to the Spark:
ssh-copy-id basu@spark-2944-dgx-userspace.tailc87c5a.ts.net
On Windows, run this in PowerShell instead:
type $env:USERPROFILE\.ssh\id_ed25519.pub | ssh basu@spark-2944-dgx-userspace.tailc87c5a.ts.net "mkdir -p ~/.ssh && cat >> ~/.ssh/authorized_keys"

For a nicer setup, use VS Code with the Remote - SSH extension and connect to basu@spark-2944-dgx-userspace.tailc87c5a.ts.net.
Run long jobs inside tmux so they keep going if your connection drops. Start a session with tmux, detach with Ctrl+B then D, and get back to it with tmux attach.
To check your usage:
spark-limits
free and nproc also show your share. htop and nvidia-smi show the whole machine.

Can't connect, or the name doesn't resolve	Check that Tailscale is on and that you accepted the invite.Then try the IP instead: `ssh basu@100.95.182.62`
"Permission denied"	Wrong password, so try again.Check that the username is `basu`.
A job suddenly stopped	It probably went over 60 GB of memory.You'll see a `[limit]` message in your terminal. Use a smaller batch size or model.
"Disk quota exceeded"	You've used your 20 GB in /tmp or /var/tmp.Delete temporary files there, and save big files in your home folder instead.
Anything else	Ask Shivam.