Apple's first MacBook Air and MacBook Pro systems based on the next-generation M2 processors aren't due to be released until July, but the first benchmark results are already here. However, newly-posted Apple M2 benchmark results show that the new processor doesn't tout explosive gains in general-purpose CPU performance, with a 10% gain in single-threaded and a 16% gain in multi-threaded work. However, according to this leaked benchmark, its GPU is significantly faster than its predecessor — up to 67% faster.
Someone with an Apple M2-based system has submitted CPU benchmarks to Primate Labs's Geekbench 5 database (where BenchLeaks discovered them). Geekbench 5 benchmarks don't reflect real-world performance across a range of applications (cryptography heavily impacts overall GB5 results), but the integer and float benchmark results can help us draw a few preliminary assumptions about what we can expect from the new Apple chips.
Header Cell - Column 0 | Apple M2 | Apple M1 | Apple M1 Pro 8C | Core i7-12800H | Core i7-11800H | Ryzen 7 5800H |
---|---|---|---|---|---|---|
General specifications | 4P, 4E, up to 3.49 GHz | 4P, 4E, up to 3.20 GHz | 6P, 2E, up to 3.22 GHz | 6P, 8E, up to 4.80 GHz | 8P, 2.30 ~ 4.60 GHz, 24MB | 16P, 3.20 ~ 4.40 GHz, 20MB |
Single-Core | Integer | 1759 | 1597 | 1616 | 1563 | 1331 | 1247 |
Single-Core | Float | 2083 | 1896 | 1896 | 1885 | 1556 | 1617 |
Single-Core | Crypto | 3021 | 2783 | 2812 | 3703 | 3784 | 3546 |
Single-Core | Score | 1919 | 1746 | 1760 | 1767 | 1521 | 1473 |
Multi-Core | Integer | 8196 | 7013 | 8592 | 11305 | 8231 | 8081 |
Multi-Core | Float | 9840 | 8624 | 10460 | 12064 | 8873 | 9239 |
Multi-Core | Crypto | 12964 | 10137 | 17028 | 8453 | 6272 | 5075 |
Multi-Core | Score | 8928 | 7653 | 9574 | 11390 | 8326 | 8305 |
Link | Link | Link | Link | Link | Link | Link |
Apple's M2 has the same CPU and GPU architectures already used by Apple's A15 SoC for smartphones. Meanwhile, since M2 leverages the performance-optimized N5P fabrication technology, it can run its high-performance Avalanche cores at up to 3.49 GHz, yielding a 10% gain over its predecessor. Apple's high-performance Avalanche cores and energy-efficient Blizzard cores are faster than previous-generation Firestorm and Icestorm cores used for the M1 generation of SoCs, but mainly because of larger clocks, larger caches, and an improved memory subsystem — not as a direct result of significant IPC gains.
In general, Apple's M2 isn't much faster than its predecessor in single-thread workloads, which is surprising. However, Apple's new Avalanche cores are significantly faster in single-thread workloads than Intel's Golden Cove cores (used for Alder Lake CPUs), and this is perhaps what matters most.
As far as multi-thread performance is concerned, the M2 is tangibly faster than the M1, with up to 16% more performance. Part of that is because the A15 Blizzard e-cores have a much more significant improvement than the Avalance p-cores, but higher clocks, enhanced caching algorithms, and a better memory subsystem all factor in. As a result, the M2 approaches Apple's M1 Pro SoC in multi-threaded workloads. Meanwhile, since Intel's 12th Generation Core 'Alder Lake' processors have more cores than Apple's M2, they are significantly faster in GB5's multi-threaded workloads.
But while the M2's performance in general-purpose workloads looks like a mixed bag (we will have to test ourselves before drawing any conclusions here), graphics is where the new SoC shines — our comparison above shows an impressive 67% improvement. It is because Apple has increased the number of GPU clusters in its entry-level SoC. It also uses a new GPU architecture for A15 and M2, a combination that brings a rather dramatic performance enhancement to the chip, at least based on results obtained in Geekbench 5. We can only wonder what to expect from real-world applications, but M2's new memory subsystem will be advantageous here.
Current test results should be considered preliminary and taken with a grain of salt. But generally, it looks like the M2 will not significantly improve single-thread performance. Instead, it will focus on higher performance in multi-threaded CPU workloads and bring an even more considerable boost to GPU-processed tasks.