Run Local AI on an Old Laptop

Record the OS, exact CPU, RAM, and free storage, then compare them with the runtime requirements. An old CPU can have enough RAM yet lack a required instruction set.

Connect power and clear the vents before sustained inference.

For verification, save the model and runtime versions, source input, relevant settings, and observed output together. Repeat the step while changing only one factor, and record unexpected results and untested limits as carefully as successes before applying the guidance to private or production data.

Close memory-heavy apps, use a 1B–3B Q4 model and short context, then repeat the same prompt.

The working rule for “Create a small baseline” is: Check instruction-set support, storage, cooling, and RAM together. Complete one small test with public data first, then record each changed condition and its effect before increasing scope.

Record the current version and settings before the example, then verify the expected response, file, or process afterward. Preserve the error and return to the smallest working command before adding options; this separates installation failures from input and integration failures.

ollama run gemma3:1b
ollama ps

Test short rewriting, classification, and summaries that a person can verify. Long documents and concurrent users may require newer hardware.

The working rule for “Choose realistic work” is: Judge reuse by task completion, not headline speed. Complete one small test with public data first, then record each changed condition and its effect before increasing scope.

Define completion with an observable result instead of a general impression. Repeat the same input, and if the output changes, isolate whether the model, runtime settings, or source data changed before moving to the next stage.

Extra RAM or an SSD can help when the CPU remains supported. Replacement is more rational when compatibility, battery, or cooling costs dominate.

The working rule for “Decide whether to upgrade” is: CPU-only tests can begin with a quantized 1B–3B model. Complete one small test with public data first, then record each changed condition and its effect before increasing scope.

For verification, save the model and runtime versions, source input, relevant settings, and observed output together. Repeat the step while changing only one factor, and record unexpected results and untested limits as carefully as successes before applying the guidance to private or production data.