Some utilities that I've discovered today in this reddit post:

• Pulse Diagnostics: a tool to show information about Windows. Nice-looking UI. It includes a simple view to see the Windows log and the list of drivers.

• Replicaro: a backup tool.

• AltRun: a launcher for Windows. Paires well with the amazing Everyghing search tool, which is, by far, the best search tool for Windows. It is fast and lightweight.

A quick note about estimating which AI models can be run locally:

• Can I Run AI locally is a website that lists which AI models your system can run from information gathered using browser APIs.
• LLMFit is a terminal tool that does the same thing but it's a CLI and TUI implemented in Rust.
• LLM Checker is a Node.js CLI tool that also tells you which models you can run locally, but it doesn't do it from system specs but actually tries to run the models (using Ollama) and see if they work.

Semianalysis has published a fantastic analysis of the current state of CPUs and GPUs in datacenters, including a thorough history of CPUs and GPUs in datacenters. This is my summary.

Global idea: from 2023 to 2025Q4, AI training made GPUs more important than CPUs in datacenters. However, since then, Reinforcement Learning and vibe coding are making CPUs more important again.

Also important: ARM is growing (with processors like AWS Graviton and Nvidia Grace), AMD is growing and Intel is shrinking.

A brief history of CPUs and GPUs in datacenters

1990s: origin of the modern datacenter. The success of the PC made it possible to sustitute costly workstations and IBM mainframes. Intel creates chips for servers: Pentium Pro (1995) and Xeon (1998), with more cache than desktop chips.

2000s: dot com era. GHz race finished because of the end of Dennard scaling: around 2004 decreasing transistor size stopped proportionally improving the power consumption, so frequency could not increase further. Multicore era starts, with more cores, SMT (Simultaneous Multithreading) and multisocket systems instead of higher clock speeds …

The Linux utility strace is essential for diagnosing process–kernel interactions, but its default output is often unusable. The key to effective debugging is using a specific set of flags that transform raw system call data into a structured, time‑stamped, and annotated log.

According to Avikam Rozenfeld in this presentation, here is the essential command template, followed by a breakdown of why each flag is critical:

strace -f -s 256 -o trace.log -tt -T -y <your_command_here>

Flags and why they matter:

• -f — Follow children
  Purpose: Trace child processes spawned by fork/clone.
  Key benefit: Ensures you trace the entire application flow (e.g., piped commands).

• -s 256 — Increase string size
  Purpose: Increase the string output limit (default 32 bytes) to 256 bytes.
  Key benefit: Prevents truncation of file paths and data being read or written.

• -o — Output to file
  Purpose: Redirect all strace output to a specified log file (e.g., trace.log).
  Key benefit: Separates trace output from the program's standard output for easier analysis.

• -tt — Precise timestamp
  Purpose: Prefix every line …

To build a new OS, or not to build a new OS?

There's a fascinating debate in the software world about whether it still makes sense to create a new general-purpose operating system from scratch.

The Pragmatic View

On one side, you have figures like John Carmack. In a recent discussion on X, he argued that building a new OS is often impractical. The cost, short lifespan, and developer burden rarely justify the effort, a lesson he learned from opposing Meta's custom XR OS.

The Idealistic View

On the other side is the spirit of the ultimate craftsman. This is captured perfectly in a joke by DHH during this presentation:

"People who are really serious about software should make their own operating system."

He's riffing on a famous quote by Alan Kay about hardware, but the message is clear: the ultimate challenge for a software purist is to build the whole stack.

A Middle Ground: Omarchy

Interestingly, DHH's own work offers a third path. He hasn't built an entirely new OS. Instead, he created Omarchy …