AMD Gorgon Halo 192GB vs M4 Max 128GB for local LLMs
8 reports on the AMD Gorgon Halo 192GB and 52 on the M4 Max 128GB
Which is faster for local LLMs?
On gpt-oss 117B · 5.1B active at 4-bit, the AMD Gorgon Halo 192GB has 49.0 generation tokens/s (2 runs); prompt processing 760.0 tokens/s (2 runs); the M4 Max 128GB has 85.4 generation tokens/s (one run); prompt processing 809.5 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 2.0x higher on the M4 Max 128GB (273 GB/s on the AMD Gorgon Halo 192GB, 546 on the M4 Max 128GB). Across 3 shared models with generation measurements on both cards, the AMD Gorgon Halo 192GB's median is higher on 0 and the M4 Max 128GB's on 3.
- Memory
- 192GB unified
- Memory bandwidth
- 273 GB/s
- FP16 compute
- 61.4 TFLOPS
- Reports
- 8
- Memory
- 128GB unified
- Memory bandwidth
- 546 GB/s
- FP16 compute
- 18.4 TFLOPS
- Reports
- 52
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 | AMD Gorgon Halo 192GB | M4 Max 128GB |
|---|---|---|
| gpt-oss 117B · 5.1B active4-bit | 49.02 runsPP 760.0 (2 runs)llama.cpp512 context, 1 not stated | 85.4one runPP 809.5 (one run)llama.cppcontext not stated |
| Qwen3.8 27B6-bit | 9.0one runPP 463.0 (one run)llama.cppcontext not stated | 17.2one runPP 246.3 (one run)llama.cppcontext not stated |
| Qwen3.8 27B8-bit | 8.1one runPP 466.0 (one run)llama.cppcontext not stated | 15.7one runPP 255.7 (one run)llama.cppcontext 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 | AMD Gorgon Halo 192GB | M4 Max 128GB |
|---|---|---|
| gpt-oss 117B · 5.1B active | 11.7 t/sQ8_0 | 38.6 t/sQ5_K_S |
| Qwen3.8 27B | 2.8 t/sFP16 | 6.7 t/sFP16 |
| Qwen3.8 125B · 6B active | 9.5 t/sQ8_0 | 30.7 t/sQ5_K_S |
| Qwen3.6 35B · 3B active | 10.0 t/sFP16 | 21.2 t/sFP16 |
| DeepSeek V4 Flash 284B · 13B active | 12.5 t/sQ4_K_M | 48.3 t/sQ2_K |
| Qwen3.6 27B | 2.8 t/sFP16 | 6.7 t/sFP16 |
| Gemma 4 26B · 4B active | 7.5 t/sFP16 | 18.2 t/sFP16 |
| DeepSeek V4.1 Flash 552B · 16B active | 16.7 t/sQ2_K | doesn't fit at these settings |
Change the context, RAM or model in the calculator: AMD Gorgon Halo 192GB · M4 Max 128GB
Every report on each card, with its source: AMD Gorgon Halo 192GB · M4 Max 128GB