Thunder Compute: A Student's Dream Alternative to Google Colab 🚀

The Student-Friendly Secret I Wish I Knew Earlier

Hey there, fellow code wranglers! Let me share something that would have saved me so much headache during my ML projects - Thunder Compute gives you a whopping $20 credit to start! 💰

Why is this a big deal? Well, do the math:

$20 credit ÷ $0.57/hour = ~35 hours of A100 GPU time

Bring Your Own Editor - A Game Changer

One thing that always bugged me about Colab was being stuck in their notebook interface. With Thunder Compute, you can use whatever text editor makes you happy:

• VSCode: My personal go-to for most projects

• Cursor: This is where things get interesting!

The Cursor Advantage

If you haven't tried Cursor yet, it's basically VSCode with AI superpowers baked in. Think of it as having a pair programmer who never gets tired or asks for coffee breaks! 😄

The workflow is something like:

1. Spin up your Thunder Compute instance

2. Connect Cursor to it via SSH (just like I explained for VSCode)

3. Start coding with AI assistance right on your GPU-powered machine

This combo is absolutely killer for speeding up development. Imagine writing a complex PyTorch model and having the AI suggest optimizations while leveraging the raw power of an A100 GPU!

A Day in the Life Example

Let me walk you through how this actually looks in practice:

1. Morning: I fire up Thunder Compute (tnr create) while sipping coffee

2. Setup: Connect instance to my cursor code editor

3. Development: Use Cursor's AI features to help debug my transformer architecture

4. Training: Run experiments on the A100 GPU while I take notes

5. Evening: Still have 99% of my credits left because I'm only paying for what I use!

Quick Cost Comparison

The Goat Guide:

To make things smoother

This is a one-time thing to make your experience smoother

Open VSCode and install the "Remote - SSH" extension from the marketplace.

To ensure you have all necessary extensions for AI/ML development, configure your default remote extensions:

1. Open Command Palette (Cmd+Shift+P or Ctrl+Shift+P)

2. Type "Preferences: Open User Settings (JSON)"

3. Add the following configuration:

"remote.SSH.defaultExtensions": [
    "ms-toolsai.jupyter-keymap",
    "ms-toolsai.jupyter",
    "ms-python.python"
]

This is to install all the necessary extensions on the VPC.

Setup your thunder compute account

pip install tnr #one time thing 
tnr login #one time thing 
# Get your $20 credit activated (you will be needed to enter cc info)
#enter you api token

Getting Started in 2 Minutes

Want to try this setup yourself? Here's the super-quick version:

Open a terminal

tnr create # you can use other gpus with --gpu flag (Read Docs)
tns status --no-wait #(to check if the instance is up)
tnr connect 0 #connects to instance with id 0
# In another terminal, open Cursor and connect via SSH
#Once connected to the vpc, run this command in 
pip3 install ipykernel

You can close the terminal now

Supercharge your jupyter book with ai-chat/agents

1. In VSCode/Cursor , navigate to the Remote-SSH extension in the sidebar

2. Select your Thunder Compute instance (e.g., "tnr-0")

   

3. Choose "Open in New Window."

4. When prompted, click "Open Folder" and select /home/ubuntu as your working directory

   

5. Create an ipynb file and make sure to select the Python kernel. You can simply drag and drop the files to use them—thanks to SSH magic. You can leave the model training, close the code editor, and go have some fun.

   

Pro Tip

Once you are done using it, run tnr stop to stop the instance while saving your data; this will only cost you storage, which is minimal. You can always restart it with tnr start. If you use tnr delete [instanceid], it will delete your instance along with your data.

The Bottom Line

If you're a student or just learning ML, this combination of Thunder Compute + Cursor (or your editor of choice) is honestly a no-brainer. The $20 credit stretching to 39 hours means you can experiment freely without watching the clock tick down your wallet.

Plus, working in your own familiar editor with AI assistance just makes the whole experience so much more pleasant than fighting with notebook cells and session timeouts in Colab.

Have you tried this combo yet? What editor are you using with your GPU instances? Let me know what's working for you! 👩‍💻👨‍💻

Thunder Compute