Est. 2002

Notes on tech, food, and everything in between.

X2Q is a personal blog that has been online since 2002 — 24 years of notes on software, Linux, payments, home labs, cooking, and the side projects that keep appearing.

Online since 2002·Author X2Q·Get in touch →

Latest posts

Jun 24, 2026
Five alternatives to pyannote: speaker diarization for phone calls (2026)
pyannote-audio 3.1 works, but requires a Hugging Face token and takes 5 seconds per call. Benchmark of five alternatives on 8 real 2-party phone calls: WeSpeaker + spectral clustering (0.2 s), multi-scale AHC (0.6 s), PLDA + spectral (0.3 s), Silero VAD + spectral (2.1 s) and cross-call gallery nearest-neighbour (0.2 s). The Silero VAD approach won at 76.1% agreement and is 2.4× faster. Surprise: all five collapsed on mono calls — pyannote stays in production.
Jun 24, 2026
Google Coral Edge TPU Benchmarks & Experiments (2026)
Five models benchmarked on the Google Coral USB Accelerator: MobileNetV1 runs in 3.9 ms (5.3× faster than CPU), EfficientNet-L runs in 31 ms (9.7× faster), and SSD MobileNetV2 object detection runs in 12 ms (3.2× faster). Largest practical speedup comes from bigger models. This post shows the numbers and explains why.
Jun 24, 2026
Setting Up the Google Coral Edge TPU on Ubuntu 24.04 (2026)
The Google Coral USB Accelerator is a $30–$60 inference chip that can run quantized TFLite models at 4 TOPS — but getting it working on Ubuntu 24.04 with kernel 6.8+ requires patching the gasket-dkms source for four kernel API breakages. This post walks through every step.
Jun 22, 2026
Apple M1 Pro vs RTX 4070 Ti for local AI — GPU matmul and LLM benchmarks (2026)
RTX 4070 Ti hits 75 TFLOPS at fp16 via tensor cores; M1 Pro MPS reaches 4.2 TFLOPS — an 18× compute gap. But on real LLM inference (minicpm-v4.6, qwen3.5, gemma4:12b via ollama), the RTX is only 3–4× faster. LLM inference at batch=1 is memory-bandwidth-bound, not compute-bound, and the M1's unified-memory architecture closes most of the gap. Apple's AMX extensions also give its CPU 2.7× better matmul throughput than an Intel i5 with the same thread count.
Jun 22, 2026
Ollama complete guide — run LLMs locally (2026)
Ollama lets you run large language models locally with a single command. This guide covers installation on Mac and Linux, how to pick the right model (with a size table), a CLI cheat sheet, the REST API with curl and Python, customisation with Modelfile and system prompts, and performance tricks that halve VRAM usage.
Jun 22, 2026
Rust vs Go vs C on Apple Silicon and Linux — benchmarks on real hardware (2026)
Four benchmarks on two machines — Apple M1 Pro (macOS/clang) and Intel i5-13400F (Linux/gcc). Rust ≈ C on the M1; on x86 Linux, Rust is 1.5× slower than gcc on recursive fib and Go is 2.9× slower. Go's matrix multiply gap is 4.7× on Linux (vs 3.3× on arm64). The surprise: Rust's concurrent sum on x86 is 2.8× faster than C pthreads — LLVM likely folds the range into a closed-form sum.
Jun 10, 2026
BIN/IIN lookup: the card prefix ranges and how to look one up (2026)
A practical reference for BIN/IIN lookups: what the issuer identification number is, the table of which prefix ranges belong to which card network, and the three ways to look a BIN up — read it off the ranges yourself, use a lookup site/API like binlist.net, or query it programmatically. Includes the 6-vs-8-digit caveat and what you can and can't tell from a prefix alone.
Jun 10, 2026
Transcribing 33,000 Danish voice logs on home GPUs — the local pipeline (2026)
Business phone calls had been recorded for 13 months — ~33,000 Danish mp3s, ~570 hours of 32 kbps phone audio. The job: transcribe everything, name the speakers, summarize per call/day/week, browsable on a site, and do it locally with no LLM API. This is the build: a benchmark of Danish ASR models, a dual-model + Claude-fusion transcription pipeline, 'phone-first' speaker identification from metadata, self-healing infrastructure across two GPUs, and Claude Code subagents as the (API-less) summary layer. Plus what it teaches about applying the same pipeline to a customer-service function.
Jun 10, 2026
Factory reset a Yealink phone — even without the admin password (2026)
You picked up a used Yealink desk phone, or inherited one with an unknown admin password. Here's how to factory reset it three ways — from the keypad, from the web UI, and (the important one) with a hardware key-hold that doesn't need the admin password at all. Plus what changed about Yealink's default password: 'admin/admin' on older firmware, a unique per-device password on newer units.
Jun 10, 2026
Update Garmin firmware and maps in 2026 — from WebUpdater to Garmin Express
The old way to update a Garmin — WebUpdater — was discontinued years ago. In 2026 you use Garmin Express (a desktop app, cable connection) for firmware and map updates on nüvi/DriveSmart units, or Garmin Connect for over-the-air updates on watches and newer handhelds. Here's the current path for each, what to do with very old nüvi units, and the hidden nüvi debug menu.

About

How it started

X2Q started back in 2002 as a plain ~cc homepage on the university server, while I was a computer-science student (stud.scient.) at Aalborg University. It was the usual student mix: Linux notes, write-ups of semester projects, and whatever I happened to be tinkering with.

Over the years it moved off the university server, through a few different engines and domains, and became x2q.net — but the habit never changed: notes to my future self, published in the open. You can still see the very first version at cs.auc.dk/~cc/ in the Internet Archive.

Notes on tech, food, and everything in between.

Five alternatives to pyannote: speaker diarization for phone calls (2026)

Google Coral Edge TPU Benchmarks & Experiments (2026)

Setting Up the Google Coral Edge TPU on Ubuntu 24.04 (2026)

Apple M1 Pro vs RTX 4070 Ti for local AI — GPU matmul and LLM benchmarks (2026)

Ollama complete guide — run LLMs locally (2026)

Rust vs Go vs C on Apple Silicon and Linux — benchmarks on real hardware (2026)

BIN/IIN lookup: the card prefix ranges and how to look one up (2026)

Transcribing 33,000 Danish voice logs on home GPUs — the local pipeline (2026)

Factory reset a Yealink phone — even without the admin password (2026)

Update Garmin firmware and maps in 2026 — from WebUpdater to Garmin Express