Field notes from the bottom of the spec sheet
Run AI on 8 GB of RAM.
No GPU. No nonsense.
Most local-AI guides quietly assume 32 GB and an RTX card. This lab runs on a 2018 ultrabook: Core i5, 8 GB of RAM, no discrete GPU. Every guide here is a test result, not a spec sheet. If it doesn't run here, we say so.
One test machine No sponsored verdicts Updated weekly
The short answer
What actually runs on 8 GB
Quantized models, CPU inference, browser closed. These are the configurations that survived testing — not the ones that merely installed.
| Model | Quant | File size | Speed (this CPU) | Verdict | Use it for |
|---|---|---|---|---|---|
| Qwen3 1.7B | Q4_K_M | 1.1 GB | ~12–18 tok/s | Runs | Chat, short drafts |
| Llama 3.2 3B | Q4_K_M | 2.0 GB | ~6–9 tok/s | Runs | General chat, summaries |
| Qwen2.5 7B | Q4_K_M | 4.7 GB | ~2–4 tok/s | Needs tweaks | Quality when you can wait |
| Gemma 3 4B | Q4_K_M | 2.5 GB | ~4–6 tok/s | Runs | Writing, explanations |
| Phi-4-mini 3.8B | Q4_K_M | 2.3 GB | ~5–7 tok/s | Runs | Code explanations |
| Whisper base (CPU) | ggml | ~150 MB | 10 min audio ≈ 7 min | Runs | Transcription |
tok/s = generated tokens per second, measured with the browser closed. Speeds are rounded ranges across five runs.
Pick a lane
Start where you are
PATH 01
“I want local AI on this laptop.”
Get a model answering questions today — then learn which sizes are actually usable at 8 GB.
- Why 7B models fit at all
- The 1B–4B sweet spot
- Ollama vs llama.cpp
PATH 02
“I need to code on a slow machine.”
A Python and VS Code setup that leaves enough memory for the actual work.
- Python without the bloat
- VS Code at 4-second startup
- AI help that doesn't freeze the editor
PATH 03
“Should I just upgrade?”
What 16 GB, a better SSD or a used mini PC actually buys you — with prices.
- 8 vs 16 GB for local AI
- Used mini PC vs cloud
- The gear that earns its place
Newest field reports
Tested this month
Can You Run an LLM on 8 GB of RAM? What Actually Works
Yes, a language model runs on 8 GB of RAM with no GPU — but only if you respect the memory math. Here is what fits, what crawls, and the exact setup I use.
The Best Small LLMs for 8 GB RAM, Tested on a CPU-Only Laptop
Qwen3, Llama 3.2, Gemma 3, Phi-4-mini and SmolLM2 compared at 8 GB: real speeds, real memory use, and which one to pick for chat, writing or code.
GGUF Quantization Explained: Q4, Q5, Q8 and What Fits in 8 GB
What quantization actually does to a language model, the difference between Q4_K_M, Q5_K_M and Q8_0, and how to pick the right one for 8 GB of RAM.
Ollama vs llama.cpp on an 8 GB Laptop: Which Is Actually Faster?
I ran the same Qwen3 and Llama models through Ollama and llama.cpp on a CPU-only laptop. The speed gap is smaller than the convenience gap — here is when to use each.
A Local Coding Assistant on an 8 GB Laptop: What's Usable, What's Not
I wired up Continue and Ollama as a private coding assistant on an 8 GB CPU-only laptop. Inline autocomplete failed; code explanation and review worked. Here is the setup.
Free Cloud GPU in 2026: Colab vs Kaggle vs Hugging Face — Real Limits
What the free tiers actually give you: GPU hours, RAM, session limits, storage and the catches. Plus the workflow I use to offload image generation and training from an 8 GB laptop.
SPIN
The rule: nothing gets published until it runs on the test bench. Every article carries the machine spec, the measured wait, and a verdict. When something is too slow to use, we write that down instead of burying it under “results may vary.”
Start here if you read nothing else
The three guides that matter
Can You Run an LLM on 8 GB of RAM? What Actually Works
Yes, a language model runs on 8 GB of RAM with no GPU — but only if you respect the memory math. Here is what fits, what crawls, and the exact setup I use.
The Best Small LLMs for 8 GB RAM, Tested on a CPU-Only Laptop
Qwen3, Llama 3.2, Gemma 3, Phi-4-mini and SmolLM2 compared at 8 GB: real speeds, real memory use, and which one to pick for chat, writing or code.
A Local Coding Assistant on an 8 GB Laptop: What's Usable, What's Not
I wired up Continue and Ollama as a private coding assistant on an 8 GB CPU-only laptop. Inline autocomplete failed; code explanation and review worked. Here is the setup.
Why this site exists
The 8 GB laptop is not e-waste.
A 2018 ultrabook with 8 GB of RAM can still write code, transcribe interviews, summarize documents and run a small language model. What it cannot do is pretend to be a workstation. This site maps the boundary precisely, so you spend money only where it changes what you can do.