Mac Mini M6 Sustains a 600MHz Emulated Pentium II in 86Box, 20% Past the M4

A hands-on review pits Apple’s new 12-core Mac Mini M6 against the base M4 as a host for 86Box, a hardware-level emulator that models a period PC down to CPU timing, chipsets, buses, and graphics chips. Because that accuracy is expensive and lands almost entirely on one host thread, the machine that wins is the one with the fastest single core that can hold its clock without throttling — precisely Apple Silicon’s strength. The reviewer treats each Mac as a ceiling rather than a score: for a fixed emulated config there is a maximum clock the host can drive at a flat 100% of target speed, and dipping below that is disqualifying because starved audio buffers produce audible dropouts in real time.

The test rig emulates a Slot 1 Pentium II (Deschutes) with 256MB RAM, a Voodoo 3 with 16MB, and Windows 98 SE, running on a custom 86Box 6.0 build whose frequency tables were extended in 50MHz steps to 800MHz. The build also adds an ARM64 JIT recompiler for Voodoo graphics, so some of the gain is better software, not just faster cores. Under a deliberately awkward load — Cinebench 2000 while Winamp plays a WAV — the base M4 holds 500MHz but hitches at 550MHz, while the M6 cleanly passes both 550MHz and 600MHz, sustaining 100% through a 7–8 minute 3DMark 2000 SE demo. That makes the M6’s top stable clock 20% higher than the M4’s. Under the hood, 86Box parks the work on two ‘super’ cores (cpu6/cpu7) running roughly 4.5–4.7GHz while the rest of the package idles.

The more interesting wrinkle is a large gap between emulated and real-world numbers: the emulated 450MHz PII scores about 61% higher than a user-reported physical 450MHz result, and the 600MHz run lands next to a real 800MHz Coppermine. The discrepancy appears even at 300MHz and 450MHz, below the reviewer’s added frequencies, so it isn’t an artifact of the overclock patch. The author points to 86Box’s own documented caveats about P6 out-of-order execution and L2 cache modeling as a likely cause and is careful not to present the scores as equivalent to real Pentium II or III performance — the takeaway is the M6’s sustained-clock headroom, not benchmark parity.