M2 Pro 32GB vs M3 Ultra 512GB for local LLMs
4 reports on the M2 Pro 32GB and 41 on the M3 Ultra 512GB
Which is faster for local LLMs?
On Qwen3.8 27B at 4-bit, the M2 Pro 32GB has 8.6 generation tokens/s (one run); prompt processing 21.9 tokens/s (one run); the M3 Ultra 512GB has 31.8 generation tokens/s (one run); prompt processing 391.2 tokens/s (one run). Those are community reports whose engines and settings differ, so the gap is a rough sign, not a controlled measure of the two cards. Memory bandwidth, which caps how fast a card can write, is 4.0x higher on the M3 Ultra 512GB (205 GB/s on the M2 Pro 32GB, 819 on the M3 Ultra 512GB).
- Memory
- 32GB unified
- Memory bandwidth
- 205 GB/s
- FP16 compute
- 6.8 TFLOPS
- Reports
- 4
- Memory
- 512GB unified
- Memory bandwidth
- 819 GB/s
- FP16 compute
- 28.4 TFLOPS
- Reports
- 41
Measured on both
Generation and prompt processing speeds from plain runs (one device, one request, no speculative decoding, the whole model in the device's memory) of the same model size at the same bit class. Each side shows the engines and reported context behind its median. Context can mean a configured window or prompt depth, so these are community comparisons with differing settings. How to read these
| Model | M2 Pro 32GB | M3 Ultra 512GB |
|---|---|---|
| Qwen3.8 27B4-bit | 8.6one runPP 21.9 (one run)llama.cpp128K context | 31.8one runPP 391.2 (one run)MLXcontext not stated |
Estimated by the calculator
The quant the calculator recommends for each card and its estimated speed at a 32,768-token context with 32 GB of system RAM, from memory bandwidth and the model's shape. Where a model was also measured above, the measurement wins.
| Model | M2 Pro 32GB | M3 Ultra 512GB |
|---|---|---|
| Qwen3.8 27B | 11.5 t/sQ4_K_S | 8.0 t/sFP16 |
| Qwen3.8 125B · 6B active | doesn't fit at these settings | 13.6 t/sFP16 |
| Qwen3.6 35B · 3B active | 41.9 t/sQ3_K_M | 28.7 t/sFP16 |
| DeepSeek V4 Flash 284B · 13B active | doesn't fit at these settings | 22.6 t/sQ8_0 |
| Qwen3.6 27B | 11.5 t/sQ4_K_S | 8.0 t/sFP16 |
| Gemma 4 26B · 4B active | 25.4 t/sQ4_K_M | 21.5 t/sFP16 |
| DeepSeek V4.1 Flash 552B · 16B active | doesn't fit at these settings | 30.0 t/sQ4_K_M |
Change the context, RAM or model in the calculator: M2 Pro 32GB · M3 Ultra 512GB
Every report on each card, with its source: M2 Pro 32GB · M3 Ultra 512GB