Apple’s M6 chip has landed, first inside the refreshed Mac mini, and the early numbers confirm this is the biggest architectural jump since the M1. The headline change is the new “Super Core” design on TSMC’s first commercial 2nm process (N2), and it shows up directly in single-core performance. Here’s what the benchmarks actually say, and how the M6 lines up against AMD’s and Intel’s current flagships.
A note on the numbers below: retail M6 units aren’t widely available yet, so these are early engineering-sample results pulled from third-party CPU benchmark databases (Geekbench Browser, cpu-monkey, etc). We’ll update this post once updated numbers are available.
Chip Specifications at a Glance
Before the benchmark numbers, here’s how the four chips actually compare on paper.
| Spec | Apple M6 | Apple M5 (base) | AMD Ryzen 9 9950X3D | Intel Core Ultra 9 285K |
|---|---|---|---|---|
| Process node | 2nm (TSMC N2) | 3nm (TSMC N3P) | 4nm (TSMC) | 3nm (TSMC N3B) |
| CPU cores / threads | 12 / 12 (2 Super + 4 Performance + 6 Efficiency) | 10 / 10 (4 Performance + 6 Efficiency) | 16 / 32 | 24 / 24 (8 Performance + 16 Efficiency) |
| Max clock | Not published | Not published | Up to 5.7GHz boost | Up to 5.7GHz boost |
| GPU | 12-core, Neural Accelerators, hardware ray tracing | 8 or 10-core, Neural Accelerators | Radeon integrated graphics (2 CUs) | Intel Arc, 4 Xe cores |
| Neural engine / NPU | Dual 16-core Neural Engine | 16-core, 42 TOPS | None (chipset-dependent) | Intel AI Boost NPU |
| Max memory | 32GB unified | 32GB unified | 128GB officially (some boards push 192–256GB unofficially) | Up to 192GB DDR5-6400 |
| Memory bandwidth | Up to 170GB/s | Up to 153.6GB/s | ~89.6GB/s (dual-channel DDR5) | ~102GB/s (dual-channel DDR5-6400) |
| L3 cache | N/A (unified memory architecture) | N/A (unified memory architecture) | 128MB (3D V-Cache) | 36MB |
| TDP / power | Not published; prior-gen Mac mini measured under 50W at full load | Not published; MacBook Air rated for a 22W thermal envelope | 170W rated (230W PPT boost limit) | 125W base / 250W max turbo |
| Socket / form factor | Soldered, Mac mini | Soldered, Mac/iPad | AM5 desktop socket | LGA1851 desktop socket |
The contrast in memory bandwidth and cache strategy explains a lot of the benchmark gap. Apple’s unified memory architecture gives every core direct, high-bandwidth access to the same pool, while AMD leans on a huge 128MB L3 cache to keep its Zen 5 cores fed, and Intel splits work across two very different core types. Three different approaches to the same problem, with very different benchmark personalities as a result.
Geekbench 6: Apple’s Single-Core Lead Gets Wider
Early engineering-sample results for the 12-core M6 put it well ahead of the M5 it succeeds. Apple’s marketing claim of “up to 40% faster CPU performance” than the M4 lines up with the multi-core numbers specifically, not single-core.
| Chip | Geekbench 6 Single-Core | Geekbench 6 Multi-Core |
|---|---|---|
| Apple M6 (12-core) | 4,501 | 21,312 |
| Apple M5 (10-core) | 4,276 | 17,862 |
| Apple M4 (10-core) | 3,777 | 14,818 |
That’s a 5% single-core and 19% multi-core gain over the M5 it succeeds, launched less than a year earlier. Against the M4 baseline that most Mac mini owners are upgrading from, the M6 is 19% faster single-core and 44% faster multi-core. The multi-core figure lines up closely with Apple’s own “40% faster CPU” claim; the single-core gain, while still solid, is more modest than the marketing framing suggests.
Cinebench 2024: The Rendering Story
Cinebench tests sustained multi-threaded rendering rather than short bursts, which makes it a better proxy for real export and render workloads.
| Chip | Cinebench 2024 Single-Core | Cinebench 2024 Multi-Core |
|---|---|---|
| Apple M6 (12-core) | 220 | 1,368 |
| Apple M5 (10-core) | 208 | 1,183 |
The M6’s multi-core Cinebench score is 16% ahead of the M5 despite only adding two cores, which points to real per-core IPC gains from the Super Core architecture rather than just more silicon.
M6 vs. AMD and Intel: Where Things Get Interesting
Here’s the base M6 against the current desktop flagships from AMD and Intel: the Ryzen 9 9950X3D and the Core Ultra 9 285K.
| Chip | GB6 Single-Core | GB6 Multi-Core | Cinebench 2024 Single | Cinebench 2024 Multi | Cores/Threads |
|---|---|---|---|---|---|
| Apple M6 (12-core) | 4,501 | 21,312 | 220 | 1,368 | 12 / 12 |
| AMD Ryzen 9 9950X3D | 3,475 | 22,548 | 139 | 2,393 | 16 / 32 |
| Intel Core Ultra 9 285K | 3,379 | 23,095 | 141 | 2,261 | 24 / 24 |
Three things to note:
- Single-core, it’s not even close. The M6 beats both x86 flagships by roughly 30% in Geekbench single-core, and by more than 55% in Cinebench single-core. That’s a fanless-capable chip in a Mac mini outrunning desktop CPUs with dedicated coolers and 125W+ power budgets.
- Multi-core is roughly a wash in Geekbench, but that’s with the M6 running 12 cores against 16 and 24 cores respectively. Per-core, Apple is still well ahead.
- Cinebench multi-core flips the script. Both x86 chips pull ahead here, because Cinebench’s rendering workload scales almost linearly with thread count, and AMD and Intel simply have more threads to throw at it. This is the clearest evidence that core count still matters for sustained rendering, even against a much more efficient architecture.
The power context makes the single-core numbers more impressive, not less. The M6 does this inside a 5-inch Mac mini with no active cooling drama, while the 285K and 9950X3D are pulling triple the wattage under full load.
GPU and Neural Engine: The Bigger Leap
CPU isn’t where Apple is making its biggest claims this generation. According to Apple, the M6 delivers up to 2x faster graphics performance and up to 4x faster AI/Neural Engine throughput compared to the M4, alongside up to 2x faster SSD speeds. Metal benchmark results aren’t fully populated yet, but if the CPU claims held up this closely to Apple’s numbers, there’s no reason to expect the GPU and NPU claims won’t as well.
Who Should Actually Care About These Numbers
Most Mac mini buyers never touch a heavily multi-threaded renderer. For everyday single-threaded and lightly-threaded work, the M6 is the fastest consumer chip you can buy right now, full stop. Buyers who do live in Cinebench-style workloads (long renders, video encoding, simulation) are the one group who should still look at a 16-core Ryzen or 24-core Core Ultra 9 instead, accepting several times the power draw for it.
FAQ
Is the M6 faster than AMD’s Ryzen 9 9950X3D?
In single-core, yes, by roughly 30% in Geekbench and over 55% in Cinebench. In multi-core Cinebench, the 9950X3D pulls ahead thanks to its higher thread count.
Is the M6 faster than Intel’s Core Ultra 9 285K?
Same pattern: the M6 wins single-core comfortably, and the 285K’s extra cores give it the edge in multi-threaded rendering benchmarks like Cinebench.
Are these final M6 benchmark numbers?
No. These are early engineering-sample results from third-party CPU databases. Final retail Geekbench Browser submissions may shift slightly once review units are widely available.
