LOCALLENS VERIFIEDFIRST-PARTY TEST
Can a 2016 HP ZBook run useful local AI with Ollama?
Deviceterra tested Ollama on a 2016 HP ZBook Studio G3 with 16 GB of RAM and a Quadro M1000M GPU. The demonstration covers Qwen 2.5 3B, Phi 3.5 3.8B, and Llama 3.2 3B on the same older Windows laptop.
Video by Deviceterra · Reviewed August 22, 2026The video remains on YouTube. Deviceterra owns this first-party test video.
What this evidence supports
This first-party demonstration supports a practical claim: an older 16 GB laptop can run small local language models through Ollama. It does not prove that every 16 GB laptop will give the same result. CPU, available memory, model file, context size, background apps, and GPU support can change the experience. Start with a 3B model and treat larger models as a separate test.
Medium to high confidenceDeviceterra performed the test and owns the test machine. LocalLens reviewed the public video, its published setup record, and the named Ollama model records. No repeatable token-speed measurement was published, so this verification covers successful operation and the stated setup, not a speed promise.
Computer setup
- HP ZBook Studio G3 from 2016
- Intel Core i7-6700HQ
- 16 GB system RAM
- NVIDIA Quadro M1000M with 4 GB VRAM
- Windows 11 Pro
Model and engine
- Ollama
- Qwen 2.5 3B
- Phi 3.5 3.8B
- Llama 3.2 3B
- Exact Ollama tags, quantization, context size, and Ollama version were not recorded
Results recorded in the video
Successful local operationThree small model families were demonstrated through Ollama on the test laptopShown in video
Generation speedNo repeatable tokens-per-second result was publishedNot confirmed
Memory useNo measured peak RAM or VRAM result was publishedNot confirmed
Go to the useful parts
What the evidence supports
- Deviceterra owns the video and the laptop used for this first-party test.
- The public video description records the CPU, RAM, GPU, operating system, engine, and three model families.
- Ollama publishes Qwen 2.5 3B and Llama 3.2 3B model records, which supports the stated model sizes and availability.
What we cannot claim
- The exact Ollama model tags and quantization were not published.
- The video does not provide a repeatable token-speed benchmark, prompt set, context size, peak RAM use, or peak VRAM use.
- This result should not be applied to every 16 GB computer without checking its CPU, free memory, operating system, and engine support.