M4 Max 64GB vs AMD Strix Halo 128GB for local LLMs
18 reports on the M4 Max 64GB and 320 on the AMD Strix Halo 128GB
Which is faster for local LLMs?
On Qwen3.6 35B · 3B active at 4-bit, the M4 Max 64GB has 76.4 generation tokens/s (one run); prompt processing 319.9 tokens/s (one run); the AMD Strix Halo 128GB has 60.0 generation tokens/s (7 runs); prompt processing 1005.3 tokens/s (7 runs). 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 2.1x higher on the M4 Max 64GB (546 GB/s on the M4 Max 64GB, 256 on the AMD Strix Halo 128GB). Across 6 shared models with generation measurements on both cards, the M4 Max 64GB's median is higher on 6 and the AMD Strix Halo 128GB's on 0.
- Memory
- 64GB unified
- Memory bandwidth
- 546 GB/s
- FP16 compute
- 18.4 TFLOPS
- Reports
- 18
- Memory
- 128GB unified
- Memory bandwidth
- 256 GB/s
- FP16 compute
- 59 TFLOPS
- Reports
- 320
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 | M4 Max 64GB | AMD Strix Halo 128GB |
|---|---|---|
| Qwen3.6 35B · 3B active4-bit | 76.4one runPP 319.9 (one run)llama.cppcontext not stated | 60.07 runsPP 1005.3 (7 runs)llama.cpp, 3 not stated30K context, 6 not stated |
| Qwen3.8 27B4-bit | 23.9one runPP 119.9 (one run)llama.cppcontext not stated | 13.14 runsPP 378.2 (3 runs)llama.cpp, 1 not stated32K context, 3 not stated |
| Gemma 4 31B4-bit | 22.4one runPP 62.7 (one run)llama.cppcontext not stated | 11.52 runsPP 559.3 (2 runs)Gufocontext not stated |
| Gemma 4 26B · 4B active4-bit | 81.7one runPP 379.6 (one run)llama.cppcontext not stated | 67.32 runsPP 4061.0 (2 runs)Gufocontext not stated |
| Qwen3.5 9B4-bit | 65.5one runPP 305.9 (one run)llama.cppcontext not stated | 31.1one runPP 819.2 (one run)engine not statedcontext not stated |
| Qwen3.8 125B · 6B active2-bit | 41.5one runPP 621.3 (one run)ds4context not stated | 28.0one runPP 250.0 (one run)engine not stated128K context |
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 | M4 Max 64GB | AMD Strix Halo 128GB |
|---|---|---|
| Qwen3.6 35B · 3B active | 42.5 t/sQ8_0 | 12.6 t/sFP16 |
| Qwen3.8 27B | 13.5 t/sQ8_0 | 3.5 t/sFP16 |
| Gemma 4 31B | 11.8 t/sQ8_0 | 3.1 t/sFP16 |
| Gemma 4 26B · 4B active | 36.4 t/sQ8_0 | 9.5 t/sFP16 |
| Qwen3.5 9B | 20.2 t/sFP16 | 10.5 t/sFP16 |
| Qwen3.8 125B · 6B active | 61.4 t/sQ2_K | 12.6 t/sQ6_K |
| DeepSeek V4 Flash 284B · 13B active | doesn't fit at these settings | 33.5 t/sQ2_K |
| Qwen3.6 27B | 13.5 t/sQ8_0 | 3.5 t/sFP16 |
| DeepSeek V4.1 Flash 552B · 16B active | doesn't fit at these settings | doesn't fit at these settings |
| GLM-5.3 320B · 18B active | doesn't fit at these settings | 11.1 t/sQ2_K |
| Qwen3.5 35B · 3B active | 42.5 t/sQ8_0 | 12.6 t/sFP16 |
| Gemma 4 8B | 22.8 t/sFP16 | 11.9 t/sFP16 |
Change the context, RAM or model in the calculator: M4 Max 64GB · AMD Strix Halo 128GB
Every report on each card, with its source: M4 Max 64GB · AMD Strix Halo 128GB