inference
How Much VRAM for a 7B Model? Quantization and Context
Size a 7B model using documented memory figures for BF16, INT8 and INT4. Account for KV cache, context length…
LM Studio Guide — Local LLM Desktop Runners, GGUF Quantization Profiles & Apple Metal GPU Acceleration
Local LLM Desktop Runners, GGUF Quantization Profiles & Apple Metal GPU Acceleration
LM Studio Guide
Independent · 2026
Independent · 2026
models
Best Local LLM for 16GB RAM in 2026: What Actually Fits
Qwen3.5-9B is the best local LLM for 16GB RAM, Gemma 4 12B the runner-up. KV cache math, fit checks, and when gpt-oss-20b is worth it.
Read the article →Latest guides
comparisons
LM Studio vs Ollama for Local LLMs: How to Actually Choose
A practical comparison of LM Studio and Ollama for local LLMs: licensing limits, OpenAI API coverage, and the…
model-formats
GGUF vs Safetensors: Differences and LM Studio Support
Compare GGUF and Safetensors for inference, training, quantization and LM Studio support, including when MLX …
hardware
LM Studio System Requirements: RAM, VRAM, GPU
LM Studio system requirements for Windows, Mac and Linux: RAM and VRAM recommendations, CPU support, GPU offl…
server-api
LM Studio on Unraid: Headless Server Setup
LM Studio has no official Unraid app. The three real options for headless GPU inference on a server, what eac…
getting-started
GGUF Quantization Levels: Q4_K_M vs Q8_0 Explained
How GGUF quantization trades model quality for memory, what Q4_K_M costs you against Q8_0, and how to match a…
Start here
Running a language model on your own machine comes down to three decisions in order: whether the hardware can hold the model, whether the model exists in a format LM Studio can open, and which quantization of it to run. These cover all three.
VRAM & GGUF Sizer
Estimate the memory a model needs before downloading it. Runs in your browser.
System requirements
The hard platform floors, and how to size a machine to the model you want.
Quantization levels
What Q4_K_M actually costs you against Q8_0, and how to choose between them.
GGUF vs safetensors
Why a model will not load, and how to find a build that will.
Serving models from a home server instead of a desktop? Read the headless and Unraid options.