Offline LLM VRAM Offloading Benchmarks
Measuring token generation speeds and memory fragmentation across quantized 7B/8B GGUF models on dedicated GPU hardware using Ollama.
Hypotheses tested, boundaries pushed, and architectures broken. This is where I document raw technical experiments before they become projects — things tested, measured, compared, broken and investigated across local AI, Linux, networking and backend systems.
Ongoing bench tests across infrastructure, local AI, Linux systems, hardware and creative tech — filter by discipline or try the terminal.
Measuring token generation speeds and memory fragmentation across quantized 7B/8B GGUF models on dedicated GPU hardware using Ollama.
Engineered a custom high-capacity backup DC supply system with automated relay cutover to keep network nodes and compute rigs online during brownouts.
A modular GNU Bash provisioning suite that configures a fresh Ubuntu installation with dotfiles, developer toolchains, Docker, and kernel sysctl parameters in under 4 minutes.
Configuring custom router firmware with QoS traffic prioritization, DNS-over-HTTPS, and lightweight network-attached storage benchmarks.
Built a low-latency multi-angle camera switching pipeline using OBS Studio, virtual loopback devices, and hardware capture cards for broadcast production.
Designed and soldered a discrete power amplifier using high-current bipolar junction transistors, passive filters, and dual-rail power regulation.
Long-form lab notes with method, measurements and conclusions.