Apple launches two new chips. The M6 marks the beginning of a new generation in the Mac mini, while the M5 Ultra, as the largest chip in the M5 series, is set to power the Mac Studio.
Apple launches the M6 and the M5 Ultra. Both chips are debuting immediately in a new desktop: the M6 in the Mac mini, and the M5 Ultra in the Mac Studio. With this move, Apple is bringing two generations to the market simultaneously. The M5 Ultra is the latest and largest addition to the M5 family, while the M6 marks the start of a new series.
The other chips in both devices are not new. The M5 Pro, found in the more expensive Mac mini, and the M5 Max, which powers the entry-level version of the Mac Studio, were already announced by Apple in March alongside the new MacBook Pro. Those specifications remain unchanged.
Apple M6
The M6 features twelve CPU cores, two more than the M4 it replaces in the Mac mini. Apple claims “the fastest single-threaded performance in the world,” without supporting this statement with a benchmark. Apple tends to live somewhat in a fantasy world regarding benchmarks, so feel free to take that statement with a grain of salt. What the company does quantify: the new chip delivers forty percent higher CPU performance than the M4.
The GPU also features twelve cores, again two more than before. Each GPU core now contains Neural Accelerators: the AI accelerators already present in the M5 Pro and M5 Max, appearing for the first time in a Mac mini. Additionally, it features a Neural Engine that Apple describes as Dual 16-core, which is said to be up to two times faster than the previous generation.
In terms of memory, the M6 starts at 16 GB, configurable up to 32 GB. Bandwidth increases to 170 GB/s, compared to 120 GB/s for the M4. Apple has said nothing about the production process. The M5 series was manufactured on 3 nm; whether the M6 deviates from this remains unclear.
32 GB is not enough to work seriously with AI models, but Apple has a solution for that. You can link several Mac minis with the M6 chip together into a single cluster.
The performance figures shared by Apple are always compared against a Mac mini with an M4 featuring ten CPU cores, ten GPU cores, and 32 GB of memory. This is a justifiable choice: a Mac mini with an M5 never existed, meaning the comparison bridges two generations. For instance, processing prompts in LM Studio is said to be 4.8 times faster, and calculations in Excel 1.5 times faster. For AI performance in general, Apple speaks of four times faster, and for graphics performance and storage, two times faster in each case. –
Apple M5 Ultra
The M5 Ultra is the most powerful chip Apple currently produces. The Ultra processor is equipped with a substantial 36 CPU cores. The chip is composed of twelve so-called supercores and 24 performance cores. The latter name remains misleading, as these are actually the efficiency cores. The GPU scales up to 80 cores, also featuring Neural Accelerators in every core.
The core counts are exactly double those of the M5 Max, which has eighteen CPU cores and up to forty GPU cores. This points to the familiar approach where Apple links two Max chips together, although the term UltraFusion is not mentioned anywhere in the announcement.
Memory scales up to 512 GB, with a bandwidth of 1.2 TB/s. According to Apple, this is fifty percent more than the previous generation. The memory has a greater impact on the AI capabilities of this chip than the neural accelerators. Apple uses a single pool of memory for both CPU and GPU here, similar to AMD’s Ryzen AI Max Pro. This massive amount of memory allows users to load AI models entirely.
The Media Engine features twice as many encoding and decoding blocks as the M5 Max, capable of simultaneously playing 33 streams of 8K ProRes 422 at 30 fps.
The predecessor to this chip is the M3 Ultra: an M4 Ultra was never released. Apple therefore compares it with the M3 Ultra and, where it is more favorable, with the M1 Ultra from 2022. Compared to the M3 Ultra, the company claims 4.3 times more AI computing power, four times faster prompt processing in LM Studio, 4.3 times faster text-to-image generation, and 3.3 times faster machine learning training in Nuke.
Outside of AI tasks, the figures are significantly more modest. Rendering in Maxon Redshift is 1.7 times faster than on an M3 Ultra, and for multithreaded CPU performance, Apple reaches 1.3 times. Those not running AI workloads are therefore primarily buying an incremental upgrade.
Finally, note that the Mac Studio does not come with a full M5 Ultra by default. The base configuration has thirty CPU cores and 64 GPU cores; the versions with 36 and 80 cores are a paid option.
Four chips, two devices
With the addition of the M6 and the M5 Ultra, the current Mac range sits at a messy intersection between two generations. The M5, M5 Pro, and M5 Max remain in the lineup, with the M5 Ultra as the new flagship. The M6 stands apart from this and will likely be followed by the rest of the series over the coming months.
Both chips are featured in the updated Mac mini and Mac Studio, which are expected to be available from September 22.