Mac Mini M6: Cyberpunk 2077 Benchmarks

The 12-Core GPU

The base M6 includes a 12-core integrated GPU sharing the system's unified memory. My 24GB configuration also uses faster memory than the 16GB base model, so the results represent this specific machine rather than every M6 Mac Mini.

The base M4 Mac Mini and the entry-level 14" MacBook Pro with the binned 11-core M3 Pro (18GB) were tested with the same game version, display output, graphics settings, and in-game benchmark.

I am also using a Windows gaming desktop as a discrete-GPU ceiling: a Ryzen 9 7945HX (16-core Zen 4), 64GB of memory, and a desktop RTX 5070. Chip shortages aside, that is a reasonably attractive modern gaming setup. It is not a like-for-like comparison with Apple Silicon. The desktop runs the Windows version of Cyberpunk 2077; the Macs will run the Metal build. Hover a device name in the table for the full spec string.

It manages 16.66fps at 1080p RT Ultra, where a desktop RTX 3050 8GB lands 19fps, the desktop card is only 6.4% ahead when comparing non-RT results, with the M6 managing 37.5fps against the 3050's 39.9fps.

Cyberpunk 2077

All Cyberpunk 2077 runs use the built-in benchmark at 1080p, with MetalFX, DLSS, and FSR disabled, frame generation off, and VSync off. The shared preset ladder is Medium, High, Ultra, and Ray Tracing: Medium, with additional Ray Tracing: Ultra runs on the M6 and Gaming Desktop below. Coloured cells are average fps. Hover a result for min / avg / max.

The whole matrix is complete. Note that the demo footage further down was captured with OBS running; the numbers in the table were not.

The M6 Mini is the first Mac desktop in this set I would actually play on at native 1080p. Medium averages 55.9 fps, and more importantly the minimum is 46.57 fps, so it never drops into the range where the mouse starts feeling like a detached mess. High holds 44.2, Ultra 37.5 with a 30.75 floor. Against the base M4 that is +73% at Medium, +63% at High, and +58% at Ultra, which is a substantial jump.

Ray tracing is still a bit of a wall. RT Medium averages 20.84 fps with a 17.95 minimum, roughly 82% faster than the M4, but still not playable. You could probably turn on MetalFX and get a smooth experience, but I'll be honest I'm not interested in those numbers, if you are let me know.

Isolated from the other machines, the generational jump looks like this:

It's worth a moment to reflect there, the uplift is biggest in ray tracing. Ray tracing gains of 82% more, despite a bump of 42% memory bandwidth points to major architectural improvements from M4 into M6 which is great to see even discounting the 2 additional cores. Realistically, 82% on top of 11.5 fps is still 20.84 fps,

On the base M4 Mini, native Medium averages 32 fps and everything above that drops under 30. Ray Tracing: Medium is 11.5 fps. The binned M3 Pro is the more comfortable laptop in this set: Medium just over 40 fps, High in the mid-30s, Ultra on the 30 fps line, and RT Medium still in the low teens. The M6 desktop chip beats the older Pro laptop chip across the board.

The Gaming Desktop is a different class of machine, and a different renderer. Raster presets land in the 111-123 fps average band at 1080p, and Ray Tracing: Medium still averages 82.5 fps. That is the discrete-GPU ceiling I wanted for these 1080p runs, not a verdict on the M6. I would not read the gap as "Apple Silicon versus RTX 5070" in any strict sense.

Native RT Ultra

Turning the M6 up to Ray Tracing: Ultra at native 1080p gives 14.23 fps minimum, 16.66 fps average, and 19.34 fps maximum, with no frame generation or upscaling. It is still well below what I would call playable, but gives us another point of reference for the GPU.

The RTX 5070 Gaming Desktop manages 56.59 fps minimum, 67.66 fps average, and 81.90 fps maximum at the same native RT Ultra settings. That is a much more comfortable result, though the desktop is running the Windows build and the M6 is using Metal.

For context, TechPowerUp’s Phantom Liberty 1080p Ultra + RT chart puts the RTX 3050 8GB at 19.0 fps and the RX 7600 8GB at 14.8 fps. The M6 average sits between those figures. These are different test sets, with the M6 using the Metal built-in benchmark, so I would not read this as equivalent GPU performance. The chart below uses a selection of the supplied results.

Telemetry

GPU power during the M6 runs sat around 14-16W with the GPU pinned at 1,735MHz and 100% active almost throughout, which is the signature of a GPU-bound workload with plenty of thermal headroom left. CPU package power is missing from these traces: powermetrics on this build reports zero watts for the CPU on every capture, so the combined figure is effectively GPU-only.

powermetrics captures for each M3 Pro preset are in public/data/ (m3pro-cb2077-1080p-*.csv). The Windows runs use the same CSV shape from a local logger (gaming-desktop-cp2077-1080p-*.csv). Combined power on that machine is CPU package power plus NVIDIA board power, not an Apple SoC package.

Every Preset, Every Machine

Beyond Average Frame Rate

Average frame rate can conceal poor pacing, input latency, and visual compromises from aggressive upscaling. The final assessment will consider how each configuration feels and looks, not just the largest number in the results table.