Why I build OpenChip

First, what is OpenChip?

OpenChip is a browser-based game which I built while I was researching the topic of how AI chips are made. I see so many companies nowadays building their own chips. Some of them are focused on inference, others on training, but it seems like this is what the cool kids are doing. I got intrigued by this topic as from what I know building a chip is a no-joke task. Running inference and training a model is something I am quite familiar with, but building a chip is not a skill at my disposal. Then my thought was “Why not build a game which people with even less knowledge than mine can play and get themselves familiar with AI infrastructure and models”. I didn’t want to build a course as I (hot take here, skip if you are a chill person) hate those people creating some junky course with pre-existing online materials and no real value, while selling it for $20 at a 90% discount from an original price of $200.

So, this is not a course but a fun way to get yourself, or somebody with even less experience, to learn how AI chips, infrastructure and models work, so you can cheat on your high-school homework or whatever you use LLM for.

Inspiration

I remembered the days (I think around 2015) when I spent my evenings on the Nand2Tetris course on Coursera. I loved it. Since a very young age I wanted to build hardware but for some reason or another I went to build software. This course was so refreshing, I was able to peek inside what gates are, how to build basic components, RAM and finally a fully working system. Additionally, I love watching Ben Eater; for me this is the physical manifestation of Nand2Tetris. Both got me excited, I learned a lot about how physical computing devices work and this helped me in my career as a software engineer. I learned that no optimisation is a small optimisation, but I guess this is a topic for a separate post.

How I build it

Reading about OpenAI building their own chip, and so many startups building and disrupting (at least on paper) the industry, is genuinely exciting but I could not explain to myself - “Is it easy now to build a chip?”. I did what every sensible person does nowadays - use an LLM to help me learn. So did I. Started by exploring what other people are doing in this medium, what their background is, where they studied and what they studied, then with natural progression I started asking the real questions “How does an AI accelerator work?”, “How does a TPU work?”, “Can we embed AI models onto a chip?” etc.

One question led to another, the more I learned the more questions popped up in my head. I wanted to systematize but I didn’t want to create a course or a multi-page-long paper. Thus I decided to build a game. It is not a AAA game or something but for me it is a way to learn something in an alternative format.

“If I play the game will I be able to build chips?”

Well if you are British and you mean fried potatoes, I hope you already can do that but if you talk about CPUs, GPUs and so on, the answer is NO. This is not the goal of the game. My hope is to inspire people to learn more about AI, how it works and genuinely have fun while satisfying their curiosity.

This game is going to be improved with time so it is not a finished project for now. If you are an expert in this area and you wish to help me build a fun game so people can learn more about this vast topic of AI, infrastructure, LLMs and more you can reach out to me.

I hope you have fun and share it with someone who will find this useful.

Cheers,
Diyan B.