M3 Ultra 256GB vs M3 Ultra 512GB for local LLMs
15 reports on the M3 Ultra 256GB and 41 on the M3 Ultra 512GB
Which is faster for local LLMs?
On Qwen3.8 125B · 6B active at 4-bit, the M3 Ultra 256GB has 25.7 generation tokens/s (one run); the M3 Ultra 512GB has 28.0 generation tokens/s (2 runs); prompt processing 558.8 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 about the same on both (819 GB/s on the M3 Ultra 256GB, 819 on the M3 Ultra 512GB). Across 3 shared models with generation measurements on both cards, the M3 Ultra 256GB's median is higher on 1 and the M3 Ultra 512GB's on 2.
- Memory
- 256GB unified
- Memory bandwidth
- 819 GB/s
- FP16 compute
- 28.4 TFLOPS
- Reports
- 15
- 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 | M3 Ultra 256GB | M3 Ultra 512GB |
|---|---|---|
| Qwen3.8 125B · 6B active4-bit | 25.7one runengine not statedcontext not stated | 28.02 runsPP 558.8 (one run)llama.cpp, oMLX250K context, 1 not stated |
| DeepSeek V4 Flash 284B · 13B active8-bit | 22.0one runPP 75.0 (one run)engine not stated1024K context | 20.0one runPP 533.0 (one run)oMLXcontext not stated |
| Qwen3.8 27B4-bit | 14.0one runPP 93.1 (one run)Ollamacontext not stated | 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 | M3 Ultra 256GB | M3 Ultra 512GB |
|---|---|---|
| Qwen3.8 125B · 6B active | 27.3 t/sQ8_0 | 13.6 t/sFP16 |
| DeepSeek V4 Flash 284B · 13B active | 36.0 t/sQ4_K_M | 22.6 t/sQ8_0 |
| Qwen3.8 27B | 8.0 t/sFP16 | 8.0 t/sFP16 |
| Qwen3.6 35B · 3B active | 28.7 t/sFP16 | 28.7 t/sFP16 |
| Qwen3.6 27B | 8.0 t/sFP16 | 8.0 t/sFP16 |
| Gemma 4 26B · 4B active | 21.5 t/sFP16 | 21.5 t/sFP16 |
| DeepSeek V4.1 Flash 552B · 16B active | 48.0 t/sQ2_K | 30.0 t/sQ4_K_M |
| GLM-5.3 320B · 18B active | 19.1 t/sQ3_K_M | 7.4 t/sQ8_0 |
| Qwen3.5 9B | 23.9 t/sFP16 | 23.9 t/sFP16 |
Change the context, RAM or model in the calculator: M3 Ultra 256GB · M3 Ultra 512GB
Every report on each card, with its source: M3 Ultra 256GB · M3 Ultra 512GB