DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to check local LLM speed before buying hardware with Llmspeed

Pick a GPU or Mac and model to see the streamed speed and whether it fits in memory.

Short answer: Open https://llmspeed.dev in a browser; then pick a GPU or Mac from the list and choose a model; check the memory fit result for each quantization: fits, needs partial offload to system RAM, or won't run. All steps.

  • Top pick
  • Money
  • Beginner
  • 15 minutes
  • From the README
  • Any agent

Who it is forA buyer or developer weighing a GPU or Mac purchase who wants to run a local model and needs to know if it will feel fast enough.

The problemSpec sheets do not show how a model will stream, how long the first token takes, or whether a quantization fits in memory.

The outcomeA clear view of tokens per second, time to first token and memory fit for the exact hardware and model, with a trust label on each estimate.

What you need first

  • A web browser to use llmspeed.dev
  • For contributing a benchmark: a machine with llama.cpp and a .gguf model file
  • For local development: npm

Steps

  1. Open https://llmspeed.dev in a browser.
  2. Pick a GPU or Mac from the list and choose a model.
  3. Check the memory fit result for each quantization: fits, needs partial offload to system RAM, or won't run.
  4. Watch the answer stream and note the wait before the first token for your typical prompt length.
  5. Check the Measured, Derived or Projected label to see how much to trust the number.

Prompt for your agent

Using the llmspeed repo (github.com/nicorobinreal/llmspeed) at https://llmspeed.dev, check the simulated speed and memory fit for <model name> on <hardware name>. Expected output: tokens per second, time to first token, and fit status for each quantization. Verify that the trust label (Measured, Derived or Projected) is shown next to the speed figure.

Caveat. Llmspeed gives estimates rather than guarantees, and drivers, engine versions and thermals can move real numbers by 10 to 30%.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for nicorobinreal/llmspeed and "Check local LLM speed before buying hardware"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer