<< Back to Home Page

John IV's MAME Benchmarks

I've been benchmarking MAME since I got involved with the project in 1997 while on the MAME and MAMEUI teams.

My current i7‑14700K @5.8GHz replaces my 2021 i7‑12700K. The 14700K provides a two‑generation jump and was a simple drop-in while we wait for Nova Lake; it's providing about a 13% uptick.

Each game is run for an emulated 90 seconds (mame gamename -bench 90) which normally gets to gameplay in its demos. Some need to be set up beforehand and/or use CHDs; see the legend. Games have been chosen as historical performance pain points for MAME (analogue sound, DRCs, netlists, etc.) as it's matured, and others are just favorites. The runs mainly function to show PC power progress over time, though they often expose performance dips and increases in released code. You'll typically want a bench result of over 200% to ensure full‑speed through actual gameplay.

Numbered tests are run with the official builds from MAMEDEV.org. The results are percentages of "full speed" in the emulation. The Delta Δ column results are the nightly build's speed vs. the prior release.

My front-end IV/Play (see homepage) now has the benchmark built in; hit F9 while in the game list or F1 and use the button to kick off the benchmark suite.

A note on the various averaging methods. Previously I had just done arithmetic means; those are retained for historical comparisons. A more valid comparison nowadays is the geometric mean. This method prevents a single game with a massive percentage gain from skewing the results (pacman, robotron), giving a more accurate and realistic measure of the typical performance improvement across all games on the PCs under test.

Nightly results: MAME 0.289 nightly build mame0289-1942-gab8a0bb6888 (built 11/1/2026), benchmarked 11/1/2026.

Historical MAME Benchmark Results (2017 to Prior Year)

Pre-2017 Historical Benchmarks

Keywords: IV/Play, MAME, MAME32, MAME32QA, MAMEUI, MAMEUI64, MAMEUI.info, Benchmarking, Benchmarks, John IV