Free tools Windows power users keep installed
One-click scans. No signup required.
Fix slow local-model suggestions by finding whether the wait comes from loading, the first token, or ongoing generation; fix inaccurate suggestions by checking the model, instructions, and prompt before tuning generation settings. Start with a repeatable test, verify which hardware is actually doing the work, and change one setting at a time. The right fix depends on your runtime, model, and device—there is no single setting that guarantees faster or better writing.
Identify where the slowdown occurs
“Slow” can describe several different delays, and each points to a different cause:
- Model loading: the application takes a long time to make the model ready.
- Time to first token: the model is loaded, but it pauses before starting its response.
- Token generation: the response starts promptly but proceeds slowly.
- Long-document slowdown: short prompts are responsive, but larger inputs are not.
Test with the same model and a short, fixed writing prompt. Change only one setting between runs and compare the same task. This makes it easier to identify what helped without assuming a speed result that may not apply to your computer.
Check whether the model is using your intended hardware
Do not assume that installing a GPU or selecting a GPU option means the whole model is running on it. Allocation depends on the runtime, model, and available memory. A CPU/GPU split is worth investigating, but it does not by itself prove the cause of a slowdown.
#1 Best Overall
Ollama
With the model loaded, run ollama ps and inspect the Processor column. Ollama’s FAQ explains how it reports GPU, CPU, or split allocation.
llama.cpp
Check the startup diagnostics for GPU offload information. The llama.cpp token-generation performance guide describes the relevant output and CPU troubleshooting.
LM Studio
Review the model’s load configuration and GPU settings. The available controls and labels can vary by version; LM Studio documents them in Load a model.
Reduce unnecessary context and check memory pressure
Context is the text the model can consider for a request, including the prompt and relevant conversation. A very large context is not automatically better for a short editing task: it can increase memory use and reduce performance. Try a context size suited to the actual input, then compare results while keeping the model and prompt unchanged.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Ollama documents context-length configuration in its FAQ. LM Studio exposes context length among its model load options and in its load API documentation. In the llama.cpp OpenVINO backend, the OpenVINO guide warns that a very large resolved default context can reduce performance and describes passing an explicit -c value. Do not copy an example value without checking what the prompt and model need.
Ollama’s FAQ also documents KV-cache configuration. It identifies f16 as the default cache type and says cache quantization can reduce memory when Flash Attention is enabled. This is a memory-management option, not a way to guarantee better writing suggestions.
Tune CPU threads only for the runtime you use
If token generation is unusually slow in llama.cpp, its performance guide suggests testing -t 1. If one thread helps, increase the count cautiously, starting low and adjusting around the number of physical CPU cores until performance stops improving; then reduce it. Too many threads can oversaturate the CPU.
That advice is specific to llama.cpp. Do not apply its command-line flags to Ollama, LM Studio, or another runtime unless that runtime’s documentation supports them.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsFix inaccurate suggestions by checking the model and task first
Before changing sampling controls, confirm that the intended model is selected and that the application is using the expected prompt template and instructions. A useful editing prompt states the task, includes the text to edit, and sets constraints such as “preserve the meaning” or “return only suggested edits.” If suggestions remain unreliable, make one controlled change at a time and compare outputs on the same passage.
LM Studio documents inference settings including temperature, maxTokens, and topP, as well as context and GPU options, in its model configuration guide. These are controls to test, not universal accuracy fixes: the documentation reviewed does not establish one setting that reliably improves every model’s writing judgment.
Tell cold-start delay apart from ongoing generation
A slow first response does not always mean that every token will be slow. In the llama.cpp OpenVINO backend, the first inference token can take longer while the runtime converts the model to an OpenVINO graph; later tokens and runs are faster. This explanation is specific to that backend, so do not assume it accounts for delays in other runtimes.
Compare runtimes or hardware only after diagnosis
If you are considering a different runtime or a hardware change, first establish a repeatable baseline. Compare the same model and quantization, prompt, context, and device. Useful measures include time to first token, generation rate, memory use, usable context, backend support, and output quality on a small editing task. Record the conditions alongside any measurements; results from one machine are not a universal speed claim.
The llama.cpp OpenVINO guide says accuracy validation and performance optimization are still in progress and notes that CPU, GPU, and NPU tool coverage is not uniform. The official materials cited here provide diagnostics and configuration options, not a controlled cross-runtime leaderboard. They also do not identify a universal RAM or GPU upgrade. Confirm the specific device, model, measured allocation, and compatibility before spending money.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
