40 articles · newest first
A 24 GB card runs Qwen3.8 27B at Q4 or Q5. Q8 needs 32 GB. The surprise is the context cache: 128K tokens costs about 8 GB, a quarter of what an older 32B model charges.
Voltage · 1d ago · 9 min
A 128GB Strix Halo mini PC and a 128GB Mac Studio give an AI model about the same 96GB to work with. One costs $3,449 at Framework's US store, the other $5,099 at Apple's. Here is what the extra $1,650 buys, and what it doesn't.
Desktop · 2d ago · 13 min
32GB does not unlock 70B models. It buys better quants and a much longer memory for the 27B to 35B class. Five picks that fit a 5090, with the math shown.
Chiptune · 4d ago · 10 min
One person on one machine, or a team hitting one server? That single question decides it. Then licence, GPU and Mac support, and the API your app talks to, all read from each project's own docs.
Tensor · 5d ago · 9 min
Seven mixture-of-experts models, from gpt-oss 20B to Qwen3 235B, split into what stays on your card and what moves to system RAM. The card barely changes. The RAM you need goes from 16 GB to 96 GB.
Voltage · 5d ago · 10 min
I mapped every Mac mini and Mac Studio Apple sells today, all 17 configs, against 13 popular local AI models. If your biggest model is 27B-class, the math says a $1,299 Mac mini holds what a $3,099 Mac Studio holds.
Voltage · 5d ago · 16 min
For local AI the decider is VRAM — and that's exactly where laptops are capped. Why a desktop (or a unified-memory mini-PC) usually wins, when a gaming laptop is still right, and how to choose.
Voltage · Sep 22 · 8 min
The quiet truth: for most home users, Qwen3 is the better default than Llama 4 Scout. Where each actually wins, the LMArena controversy, and the specific cases where Scout is worth it.
Tensor · Sep 22 · 7 min
Quantization is the only reason Scout runs at home — but MoE means you shrink all 109B, not the 17B active. The honest GGUF memory ladder, the Q4_K_M sweet spot, and how low you can safely go.
Tensor · Sep 22 · 8 min
They look like a small/large ladder, but they're built for different machines. Why Scout is the home model, why the smaller one has the bigger context, and the rare case for Maverick.
Tensor · Sep 22 · 7 min
Scout is mixture-of-experts, so you're buying for 109B in memory, not 17B active. The honest GPU ladder — why a 32GB card is the sweet spot, when 24GB works, and when unified memory beats them all.
Voltage · Sep 22 · 8 min
Llama 4 didn't dethrone Qwen3 or DeepSeek — but its MoE design and huge context make it worth understanding. What Scout and Maverick actually are, the memory trap that catches everyone, and what you really need to run it.
Tensor · Sep 22 · 10 min
Quantization is the trick that lets a 16GB card run models it has no business running. Here's how GGUF quant levels work, which one I default to, the memory math, and the honest quality trade — so you get the most model your VRAM can hold.
Tensor · Sep 22 · 9 min
12GB of VRAM at ~$249, and the one card the 2026 price surge barely touched. Superb budget hardware for local LLMs — if you can live with Intel's software tax. My honest take on who should buy one.
Voltage · Sep 22 · 8 min
Qwen3 isn't one model — it's a family from 4B to 235B, and the right GPU depends entirely on which you run. My VRAM-first map of every variant, the MoE trap that trips people up, and the used card I'd hand most people.
Tensor · Sep 22 · 8 min
at one person typing, all three are basically tied. add a second and the gap turns into a cliff
Tensor · Aug 22 · 7 min
Day-zero support for the new Qwen3.8 27B just landed on two AMD boxes — one needs a graphics card, one doesn't — and the tokens-per-second numbers are real.
Tensor · Aug 18 · 6 min
Android Police ran a real local model on a real phone as a daily driver. it's not a GPU replacement — but it covers more than the GPU-shopping conversation usually admits.
Tensor · Aug 16 · 6 min
An Android app called RikkaHub Agent turns a phone into a local-LLM-powered agent that can write code, run Linux commands, and compile software on its own — as long as you keep saying yes.
Kernel · Aug 9 · 8 min
A hobbyist just proved a genuine language model can run entirely on an $8 microcontroller — no cloud, no GPU, no internet connection required.
Voltage · Aug 6 · 6 min
Phi-4 Mini, Gemma 3, and Llama 3.2 all advertise roughly the same 128K context window. Benchmark data says that number means something very different for each one.
Tensor · Aug 4 · 7 min
Four real paths to running a 70B-parameter model on your own hardware or by the hour — priced out with actual 2026 numbers.
Tensor · Aug 4 · 8 min
A book-sized box with 128GB of unified memory can run models no single consumer GPU can hold. The honest hub for the whole category — Strix Halo, Mac Studio and DGX Spark — and when a GPU still wins.
Voltage · Aug 4 · 10 min
Uncensored local models answer without the refusals and lectures of cloud chatbots. How the two kinds are made, how to run one in Ollama on an 8 GB card, and the rules that still apply after the filter is gone.
Tensor · Aug 3 · 8 min
Llama is the open model that kicked off the whole local-AI movement, and running it yourself is genuinely one command. Here's how, and which size fits your GPU.
Tensor · Aug 3 · 10 min
Mistral's models are famously efficient — they punch above their size, so you get a lot on modest hardware. Here's how to run Mistral locally in one command, by size.
Tensor · Aug 3 · 10 min
Qwen3 is arguably the best open model you can run at home right now. Getting it running is genuinely one command. Here's exactly how, and which size fits your GPU.
Tensor · Aug 1 · 10 min
No cloud, no Wi-Fi, no GPU — just a cheap microcontroller quietly telling itself bedtime stories, one token at a time.
Tensor · Jul 31 · 6 min
VRAM is the single number that decides which AI models you can run. Here's the simple rule of thumb, a size-by-size table, and a calculator to check any model against your card.
Voltage · Jul 30 · 9 min
Local AI is private, free-per-use, and offline — but the cloud is more capable and needs no hardware. Here's a straight answer on when running your own AI is worth it and when it isn't.
Voltage · Jul 30 · 9 min
Both are free, both run the same open models, and since mid-2025 both have a desktop app. The real split now: LM Studio if you want to browse, tweak and chat; Ollama if you want a model that other apps can call. Here's the full comparison, including speed.
Voltage · Jul 30 · 10 min
Gemma 3 comes in sizes from tiny to 27B, and each has a very different hardware requirement. Here's exactly what you need to run each one — and which fits the card you already have.
Tensor · Jul 30 · 9 min
Gaming laptops have fast GPUs, but they cap out at 16GB of VRAM. A MacBook Pro's unified memory goes to 128GB. For running AI models on the go, memory wins.
Tensor · Jul 30 · 9 min
You don't need a $3,000 rig to run real local AI. For about a grand you can build a machine that runs 7B models fast and 14B models comfortably. Here's the exact parts list.
Voltage · Jul 30 · 9 min
Everyone talks about the GPU price, but a local AI rig has a running cost too. It's smaller than you fear — but if you're in Europe, it's not nothing. Here's the real math.
Voltage · Jul 28 · 9 min
Both are excellent open coding models that run on a single card. But they're tuned differently, and the right one depends on how you code. Here's the honest comparison.
Tensor · Jul 28 · 9 min
8GB is the entry floor for local AI, and plenty of capable models run on it if you pick the right size and quantization. Here's what actually works, and the trap to avoid.
Tensor · Jul 28 · 9 min
Both run local models silently and efficiently, but the gap between them for AI isn't the chip — it's the memory bandwidth and how much you can buy. Here's which Mac Mini to get.
Tensor · Jul 28 · 9 min
It's slow, it's old, and it has more VRAM than cards costing twice as much. For a first local-AI GPU, the humble 3060 12GB is still the smart-money entry point.
Voltage · Jul 28 · 9 min
Everyone obsesses over VRAM, but system RAM matters too — especially the moment you try to run a model bigger than your GPU. Here's the real number for every kind of local-AI setup.
Tensor · Jul 28 · 9 min