Will that model actually run on your machine?

Crowdsourced hardware × model × runtime reports plus a deterministic memory-fit calculator. No account needed to browse or check your rig.

Your GPU or chip

Base model

Partial offload — slower

38.8 GB needed of 48 GB unified · ~36 GB usable for models

Open full calculator with these

Browse reports

See what the community has already tested.

Check a rig you don't own yet

Deterministic memory math — no account needed.

Share your numbers

Submit a report and help verify a bucket.

Recently verified

Apple M4 Pro × gpt-oss-20b

41.0 tok/s via Ollama