Consumercarts is reader-supported. Learn more
Apple M2 Ultra vs Apple M3 Ultra: Comparison
Apple M2 Ultra and Apple M3 Ultra differ on cache: Apple M2 Ultra at 48MB L2 + 8MB L1, Apple M3 Ultra at 72MB L2 + 8MB L1, which puts Apple M3 Ultra ahead on that measure. They match on 8 other specifications, including product line, form factor and market segment. In Blender Cycles (CPU) the Apple M3 Ultra scores 716 against 526 for the Apple M2 Ultra, a lead of 27%.
Specs
| Specification |  |  |
|---|---|---|
| Brand | Apple | Apple |
| Product Line | Other | Other |
| Form Factor | Desktop | Desktop |
| Model Number | M2 Ultra | M3 Ultra |
| Core/Thread Count | 24 Core / 24 Thread (16 Performance + 8 Efficiency) | 32 Core / 32 Thread (24 Performance + 8 Efficiency) |
| Release Year | 2023 | 2025 |
| TDP | 60 W | 80 W |
| Market Segment | Desktop | Desktop |
| Cache | 48MB L2 + 8MB L1 | 72MB L2 + 8MB L1 |
| Socket Status | Mature | End of Life |
| Process Node | 5 nm | 3 nm |
| Performance Cores | 16 | 24 |
| Efficiency Cores | 8 | 8 |
| SMT Enabled | false | false |
| P-Core Boost Clock | 3.49 | 4.05 |
| Base Power | 60 W | 80 W |
| Memory Type | LPDDR5X | On-Package LPDDR5X |
| Max Memory Capacity | 192 | 512 |
| ECC Support | false | false |
| Integrated Graphics | Yes | Yes |
| Retail Availability | Soldered Only | Soldered Only |
| Series | Apple Silicon | Apple Silicon |
| CPU Process Node | 5 Nanometers | 3 Nanometers |
| Architecture | Apple Silicon, Second Generation | Apple Silicon, Third Generation |
| Cores | 24 | 32 |
| Threads | 24 | 32 |
Benchmarks
Measured results, not specifications. Percentages are the gap to the leading product in each test.
Blender Cycles (CPU)
Blender 4.5, Blender Open Data (CC0). Median of monster + junkshop + classroom, combined samples/min.
Blender renders 3D scenes by tracing light through them, which loads every core at once. The score measures sustained multi-core throughput, so a chip that wins here can still lose at gaming, where one or two fast cores matter more. Figures come from Blender's Open Data project, which places no restriction on reuse, and we publish the median of every submission for that chip on the same Blender version.