A September 22, 2026 review tests retro PC emulation on a 12-core Mac Mini M6 with 24GB of memory and compares it with a base Mac Mini M4 using 10 CPU cores and 16GB. The software is 86Box, which models the timing of an old PC together with its chipset, buses, graphics, sound and storage hardware. Most of that workload runs on one host thread, so sustained single-core speed matters more than the total core count.

The test uses a modified 86Box 6.0 build based on the same commit as release build 9001. Version 6.0 was released on May 31, 2026. Its ARM improvements include faster CPU emulation and an ARM64 just-in-time recompiler for Voodoo graphics. The latter matters because the emulated Windows 98 SE system uses a Voodoo 3 with 16MB of video memory.

The emulated machine has a Slot 1 motherboard, a Deschutes Pentium II, 256MB of RAM and a two-thread Voodoo 3 adapter. The CPU clock is tested in 50 MHz steps. A custom frequency-table patch extends the range from 500 MHz to 800 MHz, scales memory and cache timings to preserve similar access latency, and keeps the AT bus at 8.33 MHz. The rest of the emulation code remains unchanged.

Each run combines the Cinebench 2000 CPU test with Winamp 2.76 playing a 16-bit, 44,100 Hz PCM WAV file. The audio stream exposes timing problems immediately. A run passes only when 86Box stays at 100% emulation speed throughout and produces no audible dropout.

The M4 remains stable at 500 MHz. It begins to hitch at 550 MHz, especially during the 3DMark 2000 SE demo. The M6 passes 550 MHz and 600 MHz. At 600 MHz, it maintains 100% speed through two Cinebench 2000 runs and a 7 to 8 minute 3DMark 2000 SE demo. Its stable ceiling is 20% higher than the M4 result.

Cinebench 2000 scores for the M6 are 4.62 at 300 MHz, 5.34 at 350 MHz, 6.16 at 400 MHz, 7.02 at 450 MHz, 7.73 at 500 MHz, 8.54 at 550 MHz, 9.28 at 600 MHz and 10.08 at 650 MHz. Both systems pass through 500 MHz. The M6 passes at 550 MHz and 600 MHz, while the M4 fails. Both fail at 650 MHz. The 650 MHz M6 run completed the render but produced one or two audible underruns, so the review keeps 600 MHz as the official result. Neither machine can sustain the added 800 MHz test point.

The scores run well above several period hardware results collected in an Ars Technica forum thread. Reported figures include 2.38 for a Pentium II 300 MHz, 4.35 for a Pentium II 450 MHz, 7.52 for a Celeron 800 MHz, 8.03 for a Celeron 533 MHz overclocked to 760 MHz, 9.25 for a Pentium III Coppermine 800 MHz and 7.66 for an Athlon Classic 600 MHz. The emulated Pentium II scores 7.02 at 450 MHz and 9.28 at 600 MHz. The first figure is about 61% above the reported physical Pentium II 450 MHz result, while the second nearly matches the Pentium III 800 MHz result.

Those comparisons are contextual because the physical systems used different memory, operating systems and other hardware. The review does not treat the scores as evidence of equivalent performance across software. Possible factors include memory bandwidth and cache timing. 86Box version 3.0 notes that P6 emulation was not fully accurate because out-of-order execution and L2 cache behaviour are complex. Deschutes timings were tuned to approach real hardware.

Telemetry shows 86Box using two performance cores, identified as cpu6 and cpu7. On the M6, they average about 4,483 MHz during the 450 MHz run and 4,710 MHz during the 600 MHz run, with a peak of 4,788 MHz. Per-core activity rises from about 41% at 450 MHz to 45% to 49% at 600 MHz. Total package usage stays below 26%. The M4 keeps its active cores near 3.7 GHz, which likely explains its lower ceiling.

The review describes the base M4 as a stable 500 MHz Pentium II machine and the M6 as a stable 600 MHz system with working audio and 3DMark playback. It calls the M6 an unusually capable small desktop for this task at a price below €1,600. Power and thermal measurements remain open because the M6 build of powermetrics reports zero CPU package power. The author also plans to test Windows XP emulation in 86Box later. An earlier 450 MHz M6 run was recorded while OBS composited and encoded 720p30 H.264, but that video documents the earlier result, not the new 600 MHz pass.