Apple unveils an M2 Max with a 12-core CPU, 30-core GPU, and support for 96GB of RAM and M2 Ultra with 24-core CPU, a 60-core GPU, and support for 192GB of RAM
20 billion more than M1 Ultra -800GB/s of system memory bandwidth and -supports 50% more memory capacity, 192GB -24-core CPU is 20% faster -76-core GPU up to 30%... https://twitter.com/... @iamdcinvestor : probably the most interesting wrinkle in all of this no one else has chips close to what the Apple headset can do https://twitter.com/... Brian Roemmele / @brianroemmele : The new Apple M2 Ultra chip will allow me to build AI models that are far larger than the memory of current GPUs. This may be the massive game changer for Private and Personal local AI. Love it! [image] Mark Hachman / @markhachman : M1 Max was 57 billion transistors. M1 Ultra doubled that, essentially. Any guesses on how big the M2 Ultra is?
Context & Ripple Effects
Apple’s Ultra line began with two M1 Max dies joined into one chip, establishing a high-end Apple-silicon tier centered on large unified memory and memory bandwidth. Reporting days before this launch had already pointed to a 60-core M2 Ultra configuration with up to 192GB of memory.
The M2 Max and M2 Ultra make that tier more capable: Apple is increasing both compute-core counts and the amount of memory available to a single system, rather than treating the Ultra as merely a faster CPU refresh.
First-order effects
- Apple’s highest-end Mac configurations gain up to 192GB of unified memory and 800GB/s of memory bandwidth with M2 Ultra, while M2 Max reaches up to 96GB.
- Developers and professional users whose workloads are constrained by GPU-accessible memory can target larger local workloads on Apple’s high-end hardware; Apple positions M2 Ultra as faster than M1 Ultra in both CPU and GPU work.
Second-order effects
- The larger memory ceiling makes memory capacity, alongside core counts, a more consequential differentiator for workstation buyers evaluating Apple silicon against discrete-accelerator systems.
- Software optimized for Apple’s integrated CPU, GPU, and shared-memory architecture has a stronger incentive to scale into Ultra-class machines, widening the performance gap between Apple’s entry and top-end Macs.
Third-order effects
- If Apple continues raising unified-memory capacity across generations, high-memory integrated systems could become a more distinct workstation category for workloads limited by data movement—the M1 Ultra’s 800GB/s-memory design is the architectural precedent.
- The move reflects a broader shift toward matching heterogeneous compute with sufficient local memory, though the practical advantage will depend on application support and workload fit.
The trend: Apple is extending its vertically integrated silicon strategy by scaling compute and unified memory together for increasingly demanding professional and local AI-adjacent workloads.