How to build a GPU

How to build a GPU
17 minutes by Jason

Transistors are tiny switches that control electrical current. They turn on or off based on a voltage signal, making them the core building block of all modern chips. Billions of them work together to store data and run calculations. Understanding how they switch is the first step to understanding hardware. Jason has built a game in which one builds GPUs from scratch.

Give your app access to Google Search with SerpApi
sponsored by SerpApi

Need Google Search data in your product? SerpApi helps you access clean, structured results from Google and other major search engines without building your own scraper. Get real-time data through a simple API, so you can power research, monitoring, analysis, and more in your app.

113 pragmatic lessons from tech lead to CTO
about 1 hour by Jack Kinsella

Jack shares 113 practical lessons from scaling a large monolithic codebase to over one million lines. It covers performance, team structure, monitoring, testing, and deployment. Key ideas include optimizing database queries, using background jobs, improving developer workflows, and strengthening observability. It emphasizes simple, pragmatic decisions and strong team practices to maintain speed, reliability, and quality as systems and teams grow.

Git commands I run before reading any code
4 minutes by Ally Piechowski

Ally reveals five git commands that show the health of an unfamiliar codebase before reading a single file. You can spot which files break most often, who actually understands the system, and whether the team is shipping steadily or putting out fires. Cross-referencing high-churn files with bug-heavy commits shows your biggest risk areas fast. The whole scan takes minutes and tells you exactly where to look first.

What is inference engineering?
36 minutes by Gergely Orosz

Open AI models are growing fast, and more companies now run their own model infrastructure instead of relying on closed APIs. Inference engineering, the work of making models run faster and cheaper in production, is becoming a key skill as a result. Core techniques include quantization, caching, speculative decoding, parallelism, and disaggregation. Investing in this area pays off once a product scales enough that off-the-shelf solutions no longer meet performance or cost needs.

USB for software developers
13 minutes by WerWolv

Writing a USB driver sounds scary, but it is actually no harder than writing a networked app. Using a library called libusb, you can communicate with a USB device entirely from user space, with no kernel code needed. Nik walks through detecting a device by its vendor and product IDs, reading its descriptors, and understanding endpoints. It wraps up by building a simple driver for Android's Fastboot protocol in just a few lines of code.

And the most popular article from the last issue was:

• The unwritten laws of software engineering