Contact

Contact

If you have questions, feedback, or correction requests, you can reach me here:

kimhoseok0502@gmail.com

I usually reply when I’m available. No promises of response time — but I do read every message.

Popular posts from this blog

DDR5 vs HBM Memory - What's Different and When Each One Wins

Image
Author: The Principle Lab DDR5 vs HBM Memory - What's Different and When Each One Wins If you have been watching chip news in 2025, you have probably seen both terms used like they are competing in the same lane. They are not. The confusing part is that both are DRAM, and both exist to feed a processor with data. So what is the real difference? It is not just "HBM is faster." The cleaner comparison is about form factor , how the interface scales, and what kind of platform the design assumes. Quick summary DDR5 is commonly delivered as client modules, and the module design includes two independent subchannels plus module-level management (PMIC/SPD/sideband). HBM3 is defined as tightly coupled to the host compute die with a distributed interface split into independent channels . DDR5 data rates are described as starting at...
Read more

The Groq LPU Architecture Overview - Deterministic Inference and SRAM-Centric Scaling

Image
Author: The Principle Lab The Groq LPU Architecture Overview - Deterministic Inference and SRAM-Centric Scaling  If you have ever wondered why some AI systems feel "snappy" while others randomly stall, you are already thinking about the right problem. For real-time inference, the pain is not just average latency. It is the unexpected spikes. Groq frames its LPU approach around predictable timing , pushing more decisions into the compiler instead of making them at runtime. This article is a hub-style map of the core ideas: the LPU mental model, deterministic inference, SRAM-first data movement, and how scaling can stay disciplined when multiple chips are involved. Quick summary Groq's LPU pitch is simple: reduce runtime surprises by planning more of the execution up front. That shifts complexity into compilation, so the hardware can run a deterministic execution ...
Read more

From Ocean Recovery to Tower Catch — How SpaceX Brings Boosters Back

Image
Author: The Principle Lab From Ocean Recovery to Tower Catch — How SpaceX Brings Boosters Back If you only see quick clips on social media, booster recovery can look like magic. One video shows spray kicking up in the ocean, another shows a rocket standing on a ship, and newer tests show a giant tower trying to grab a stage out of the air. It is easy to wonder whether someone is literally fishing Starship out of the sea. This piece walks through how SpaceX actually handles its boosters today: early ocean splashdowns, Falcon 9-style reusability with pads and drone ships, and the new tower catch experiments for Super Heavy. Along the way we will clear up the most common confusion point: what you are really looking at when a clip seems to show a booster being dragged out of the water. Quick summary if you just want the gist Here is the short version in one screen: early Starship ...
Read more