What laptop did we test, and what was the task?
The test machine is a $50 HP Notebook with an Intel Core i3-5005U (2 cores, 4 threads, 2.00 GHz), 6 GB of RAM and a SATA hard drive [1][2]. Every run was CPU-only inference: the model ran on the main processor alone, not the Intel HD Graphics 5500, and Ollama reported a 0% GPU share [1]. The task was one prompt: "Explain in about 150 words why someone might run a language model locally" [1].
How did we set up the test?
We tested Windows 10 Home 22H2 (build 19045) on September 29, 2026, then replaced it with Linux Mint 22.3 "Zena" (Cinnamon 6.6.4, kernel 6.14.0-37-generic) and tested again on October 10, 2026 [1][2]. Both ran Ollama 0.11.4, a free program that runs language models on your own computer, with tinyllama Q4_0 [1][7]. The model file's digest, its fingerprint, matched on both systems: 2644915ede352ea7bdfaff0bfac0be74c719d5d5202acb63a6fb095b52f394a4 [1].
Our unmodified bench.py script set temperature 0, seed 42 and a 200-token output cap [1][3]. A token is a small piece of text, often part of a word, and tokens per second is how many the model writes each second. We report the median of three warm runs, made with the model already loaded. Cold start is the time from an unloaded model to a finished first answer, including startup and loading from disk [1].
How much faster is Linux Mint than Windows 10 on the same hardware?
Mint was 37.2% faster in warm generation: 12.9 tokens per second against 9.4, a gain of 3.5 [1]. Cold start fell from 229.8 seconds to 19.8, 91.4% shorter. The loaded model used 0.7 GB on both systems [1].
| Measurement | Windows 10 | Linux Mint 22.3 |
|---|---|---|
| Test date | 2026-09-29 | 2026-10-10 |
| Warm speed, median of 3 runs (tokens/s) | 9.4 | 12.9 |
| Cold start to full first answer (s) | 229.8 | 19.8 |
| Loaded model memory, Ollama-reported (GB) | 0.7 | 0.7 |
| GPU share (%) | 0 | 0 |
Raw data: windows.csv and linux-mint-cinnamon-20261010-144025.csv.
Is Linux Mint faster than Windows for local AI?
On this laptop, yes: 37.2% more tokens per second with the same Ollama version, model file and settings [1]. It covers one machine and one small model, so it is evidence, not a rule.
Can an old laptop run a language model locally?
Yes, if the model is small. This 2-core laptop ran tinyllama at 12.9 tokens per second on Mint, and the loaded model needed 0.7 GB of memory [1]. For scale, a Xeon E5-2665 server in our collection reached 27.1 tokens per second with this model; the laptop managed 47.6% of that [1]. For old-laptop owners, the software setup matters too: the same hardware gained 3.5 tokens per second after the move to Mint.
Which Linux distribution is best for reviving an old laptop?
We can only speak for the one we measured. Linux Mint 22.3 Cinnamon needed one Wi-Fi driver fix on this laptop (see our install guide), then ran local AI 37.2% faster than Windows 10 [1][4]. We did not benchmark other distributions, so we offer no ranking. Mint 22.3 also comes in MATE and Xfce editions [5].
What are the limits of this test?
- It is a paired comparison of one laptop on two dates, not a lab test that isolates the operating system [1].
- Windows 10 had desktop apps and the Ollama tray app running; Mint's background activity and power profile were not recorded [1][2].
- About 167 of Windows 10's 229.8 cold-start seconds were Ollama's runner starting up, so that gap is not generation speed [1][2].
- Each system had three warm runs in one session, with no repeat session yet [1].
Where can I download Linux Mint?
Download Linux Mint 22.3 Cinnamon from mint.arynwood.com, verified against Mint's signed checksum on October 10, 2026, and follow our six-step install guide [6][4]. Linux Mint is Arynwood's official operating system because, on the hardware we test with, Mint 22.3 gave local AI 37.2% more speed than the Windows 10 it replaced [1].
What else do people ask?
Does Linux Mint need a graphics card to run local AI?
No. Our CPU-only test on an Intel Core i3-5005U reached 12.9 tokens per second with tinyllama Q4_0 on Linux Mint 22.3 [1]. Ollama 0.11.4 reported a 0% GPU share during the run.
How much memory does a small local model need?
Ollama 0.11.4 reported 0.7 GB for the loaded tinyllama Q4_0 model on both Windows 10 and Linux Mint 22.3 [1]. That is the model's loaded size, not peak system memory; the test laptop had 6 GB installed.
What is a cold start, and why was Windows 10 so slow?
A cold start is the time from an unloaded model to a finished first answer. Windows 10 took 229.8 seconds, about 167 of them spent starting Ollama's runner, while Linux Mint 22.3 took 19.8 seconds [1][2].
Can I repeat this test on my own laptop?
Yes. The bench.py script, method and raw results are public in our local-ai-benchmarks repository under CC BY 4.0 [3]. Install Ollama 0.11.4, pull tinyllama, and run bench.py in CPU mode with three warm runs and a 200-token cap [1].
Where do these numbers come from?
- Noble, L. (2026, October 10). HP Notebook: Linux Mint vs Windows 10 [Benchmark report]. local-ai-benchmarks. github.com/Arynwood-Technology/local-ai-benchmarks/blob/main/machine-tests/hp-notebook-i3-5005u/linux-mint-report-20261010.md
- Noble, L. (2026, September 29). HP Notebook: Windows vs Linux Mint Cinnamon [Machine record and Windows baseline]. local-ai-benchmarks. github.com/Arynwood-Technology/local-ai-benchmarks/blob/main/machine-tests/hp-notebook-i3-5005u/README.md
- Noble, L. (2026). Local AI benchmarks: Ollama speed and tool calling on CPU and GPU [Dataset and bench.py script, CC BY 4.0]. github.com/Arynwood-Technology/local-ai-benchmarks
- Noble, L. (2026, October 10). Replace Windows with Linux Mint: Step-by-step USB install. Arynwood Technology. mint.arynwood.com/install/
- Linux Mint. (n.d.). Download Linux Mint. linuxmint.com/download.php
- Arynwood Technology. (2026, October 10). Linux Mint 22.3 Cinnamon: verified download. mint.arynwood.com
- Ollama. (n.d.). Release v0.11.4. github.com/ollama/ollama/releases/tag/v0.11.4
How should I cite this paper?
Noble, L. (2026, October 10). Linux Mint ran a local AI model 37% faster than Windows 10 on the same $50 laptop. Arynwood Technology. https://mint.arynwood.com/linux-mint-vs-windows-local-ai/
Want help moving an old laptop to Mint, or running local AI on it? Arynwood offers setup help.