M4 24GB vs AMD Strix Halo 128GB for local LLMs
8 reports on the M4 24GB and 321 on the AMD Strix Halo 128GB
Which is faster for local LLMs?
On Qwen3.8 27B at 4-bit, the M4 24GB has 6.5 generation tokens/s (one run); prompt processing 60.8 tokens/s (one run); the AMD Strix Halo 128GB has 13.1 generation tokens/s (4 runs); prompt processing 378.2 tokens/s (3 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 AMD Strix Halo 128GB (120 GB/s on the M4 24GB, 256 on the AMD Strix Halo 128GB). Across 3 shared models with generation measurements on both cards, the M4 24GB's median is higher on 0 and the AMD Strix Halo 128GB's on 3.
- Memory
- 24GB unified
- Memory bandwidth
- 120 GB/s
- FP16 compute
- 4.3 TFLOPS
- Reports
- 8
- Memory
- 128GB unified
- Memory bandwidth
- 256 GB/s
- FP16 compute
- 59 TFLOPS
- Reports
- 321
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 24GB | AMD Strix Halo 128GB |
|---|---|---|
| Qwen3.8 27B4-bit | 6.5one runPP 60.8 (one run)llama.cppcontext not stated | 13.14 runsPP 378.2 (3 runs)llama.cpp, 1 not stated32K context, 3 not stated |
| Llama 3.2 3B4-bit | 46.0one runPP 530.0 (one run)llama.cppcontext not stated | 89.12 runsPP 1915.0 (2 runs)llama.cpp, 1 not statedcontext not stated |
| Qwen3.5 9B4-bit | 17.4one runOllama4K context | 31.1one runPP 819.2 (one run)engine not statedcontext 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 | M4 24GB | AMD Strix Halo 128GB |
|---|---|---|
| Qwen3.8 27B | 10.8 t/sQ2_K | 3.5 t/sFP16 |
| Llama 3.2 3B | 15.2 t/sFP16 | 31.6 t/sFP16 |
| Qwen3.5 9B | 10.1 t/sQ8_0 | 10.5 t/sFP16 |
| Qwen3.8 125B · 6B active | doesn't fit at these settings | 12.6 t/sQ6_K |
| Qwen3.6 35B · 3B active | 35.8 t/sQ2_K | 12.6 t/sFP16 |
| DeepSeek V4 Flash 284B · 13B active | doesn't fit at these settings | 33.5 t/sQ2_K |
| Qwen3.6 27B | 10.8 t/sQ2_K | 3.5 t/sFP16 |
| Gemma 4 26B · 4B active | 24.1 t/sQ3_K_S | 9.5 t/sFP16 |
| DeepSeek V4.1 Flash 552B · 16B active | doesn't fit at these settings | doesn't fit at these settings |
Change the context, RAM or model in the calculator: M4 24GB · AMD Strix Halo 128GB
Every report on each card, with its source: M4 24GB · AMD Strix Halo 128GB