Topic
22 articles tagged Hardware. For the wider topic, see Local & Private AI.
Virtual desktops are how a lot of regulated work happens. Whether on-device AI survives that setup, what breaks, and whether it still counts as local.
Read article →Most work laptops have no discrete GPU. Whether local AI writing is viable without one, what integrated graphics actually contributes, and when to stop worrying.
Read article →Model files are only part of it. What local AI actually consumes on a Windows laptop, where it hides, and how to reclaim it without breaking the tool.
Read article →A bigger model is not automatically the better one for rewriting. A fifteen minute test using your own writing that settles it properly.
Read article →Snapdragon laptops run local models, but not the way x86 machines do. What works today, what is still missing, and which tools quietly refuse to start.
Read article →Model size is the obvious answer and usually the wrong one. The four things that decide how long a local rewrite takes, ranked by how much they matter.
Read article →Million-token context sounds free until you run one locally. What the context window actually consumes, why it grows as you type, and what it means for a laptop.
Read article →The same local model that felt fast plugged in crawls on battery. Here is what Windows is actually doing, and the three settings that get the speed back.
Read article →Datacenter AI power draw is now a mainstream concern. The honest comparison between running a small model on your laptop and calling a frontier model, including where local loses.
Read article →What GGUF files are, how to read the cryptic names, what Q4_K_M actually means, and the three things to check before you download several gigabytes.
Read article →A tier-by-tier guide from 8GB of system RAM to a 24GB GPU, with the arithmetic behind each number and the one mistake that ruins mixture-of-experts planning.
Read article →Four-bit quantization is now the default for local models. What it costs in quality, where the floor is, and why the answer depends entirely on your task.
Read article →Nearly every large model in 2026 is a mixture of experts. What that means, why the two parameter counts matter differently, and the planning mistake it causes.
Read article →Your laptop has a neural processing unit. Ollama, llama.cpp, and LM Studio do not use it. Why the NPU story is more complicated than the marketing, and what actually runs your model.
Read article →Open weights at frontier scale sound like they put frontier AI on your desk. The arithmetic says otherwise. What the memory math actually allows, and what you should run instead.
Read article →A decision procedure that starts with your actual RAM instead of a leaderboard. Which size, which variant, which quantization, and how to know when you have picked wrong.
Read article →NVIDIA keeps shipping compressed open-weight models derived from larger ones. How pruning and distillation work, and why compressed models are the ones that reach your laptop.
Read article →Copilot+ PCs set a 40 TOPS NPU bar, but most local AI writing tools never touch the NPU. What the badge actually buys you, and what runs fine without it.
Read article →A straight answer by model size, why free RAM matters more than installed RAM, and how to work out whether your current machine can handle it before downloading anything.
Read article →Three chips, three very different jobs. Which one actually runs your local language model, why memory bandwidth beats raw compute, and what to check on your own machine.
Read article →Q4_K_M, GGUF, four-bit. What the labels on local AI models actually mean, what you lose, and which one to pick for rewriting work text.
Read article →A detailed breakdown of the computer specifications needed to run local language models smoothly, proving that you do not need a supercomputer.
Read article →