Latest AI PC guides

Practical guides, setup tutorials and buying advice.

• How Much RAM Do You Need for a Local LLM?The memory a local model needs follows a formula, not a guess. Here is the arithmetic, what each model size actually requires, and which machines clear it.
• Unified Memory vs VRAM for AIUnified memory and dedicated VRAM solve different problems. One lets you load a far larger model, the other answers much faster. Here is how to choose.
• How Much VRAM Does a 70B Model Need?A 70B model needs about 51GB at Q4 and 88GB at Q8. Here is the arithmetic, and every machine we track that can hold one without offloading.
• Can a Mini PC Run an LLM?Modern mini PCs hold models no consumer graphics card can. Here is what they run, how fast, and where their memory bandwidth becomes the limit.
• Is 128GB Enough for Local AI?128GB of system memory gives a model between 96GB and 120GB depending on architecture. Here is what that runs, and when it stops being enough.
• What Is an NPU, and Do You Need One?An NPU accelerates the AI features built into Windows. It does almost nothing for running a language model yourself. Here is the honest distinction.
• Memory Bandwidth and LLM SpeedGeneration speed tracks memory bandwidth almost linearly. Here is why, what the numbers mean in practice, and how the machines we track compare.
• Quantization Explained: Q4, Q8 and FP16Quantisation decides how much memory a model needs and how much quality it keeps. Here is what each level costs, and which to run.
• Ryzen AI Max vs RTX for Local AIA Ryzen AI Max machine holds three times the model an RTX card can. An RTX card runs what fits far faster. Here is how to decide between them.
• AI Laptop vs AI DesktopA desktop holds its clocks and takes a bigger card. A laptop throttles and runs on battery. Here is what that costs in real local AI work.