Can I run Qwen3 30B-A3B (2507) on the Apple M4?
Runs wellQ4_K_M fits in unified memory (19.9 of 22.4 GB usable by the GPU): Runs well, est. 12.8 tok/s (10.2–15.4, calibrated estimate ±20%).
32 GB unified memory · 120.0 GB/s · FP16 8.5 TFLOPS. Assumes 8k context and an f16 KV cache. Verdicts are shown for each memory size.
Verdict by memory size
The page uses the 32 GB configuration; here is every size.
- 16 GBWon't run
- 24 GBRuns slowlyest. 12.8 tok/s (9.0–16.6, theoretical estimate ±30%)
- 32 GBRuns wellest. 12.8 tok/s (10.2–15.4, calibrated estimate ±20%)
Every quant of Qwen3 30B-A3B (2507) on the Apple M4
| Quant | File size | Verdict | Speed | Memory | Runs as | Notes |
|---|---|---|---|---|---|---|
| Q2_K | 11.26 GB | Runs well | est. 17.8 tok/s14.2–21.3calibrated estimate ±20% | 12.6 / 22.4 GB | Unified memory | — |
| IQ4_XS | 16.38 GB | Runs well | est. 14.0 tok/s11.2–16.8calibrated estimate ±20% | 17.8 / 22.4 GB | Unified memory | — |
| Q4_K_Mbaseline | 18.56 GB | Runs well | est. 12.8 tok/s10.2–15.4calibrated estimate ±20% | 19.9 / 22.4 GB | Unified memory | — |
| Q8_0 | 32.48 GB | Won't run | — | Needs about 33.9 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) | — | — |
Where the memory goes at Q4_K_M
- Weights
- 18.6 GB
- KV cache
- 0.8 GB
- Compute buffer
- 0.6 GB
- Free
- 2.5 GB
- Not usable by the GPU
- 9.6 GB
Context length vs. KV cache
| Context | KV f16 | KV q8_0 | KV q4_0 |
|---|---|---|---|
| 4k | 19.5 GBRuns well est. 14.9 tok/s11.9–17.9calibrated estimate ±20% | 19.3 GBRuns well est. 16.2 tok/s13.0–19.4calibrated estimate ±20% | 19.2 GBRuns well est. 17.0 tok/s13.6–20.3calibrated estimate ±20% |
| 8k | 19.9 GBRuns well est. 12.8 tok/s10.2–15.4calibrated estimate ±20% | 19.6 GBRuns well est. 14.8 tok/s11.8–17.7calibrated estimate ±20% | 19.4 GBRuns well est. 16.1 tok/s12.9–19.3calibrated estimate ±20% |
| 16k | 20.8 GBRuns well est. 9.9 tok/s8.0–11.9calibrated estimate ±20% | 20.1 GBRuns well est. 12.5 tok/s10.0–15.1calibrated estimate ±20% | 19.7 GBRuns well est. 14.6 tok/s11.7–17.5calibrated estimate ±20% |
| 32k | 21.8 GBRuns slowly est. 6.9 tok/s4.8–8.9theoretical estimate ±30% | 21.1 GBRuns well est. 9.6 tok/s7.7–11.6calibrated estimate ±20% | 20.3 GBRuns well est. 12.3 tok/s9.8–14.8calibrated estimate ±20% |
| 64k | 25.0 GBRuns slowly est. 4.3 tok/s3.0–5.5theoretical estimate ±30% | 22.0 GBRuns slowly est. 6.6 tok/s4.6–8.6theoretical estimate ±30% | 21.5 GBRuns well est. 9.4 tok/s7.5–11.2calibrated estimate ±20% |
| 128k | 33.1 GBdoes not fit | 25.5 GBRuns slowly est. 4.0 tok/s2.8–5.3theoretical estimate ±30% | 22.2 GBRuns slowly est. 6.3 tok/s4.4–8.2theoretical estimate ±30% |
| 256k | 47.2 GBdoes not fit | 35.2 GBdoes not fit | 25.9 GBRuns slowly est. 3.8 tok/s2.7–5.0theoretical estimate ±30% |
S = Runs great · A = Runs well · B = Runs slowly · F = Won't run
At Q4_K_M the verdict stays Runs well up to 16k context with an f16 KV cache, and up to 32k with q8_0.
Estimated speed
- Generation
- est. 12.8 tok/s (10.2–15.4, calibrated estimate ±20%)
- Prompt processing
- about 383 tok/s (230–536, rough estimate ±40%)
Measured on this exact combination
No public measurement for this exact combination yet; the numbers above are estimates.
If this is not enough
Qwen3 30B-A3B (2507) on the Apple M4 Pro
Runs greatQ4_K_M: Runs great, est. 29.1 tok/s (23.3–34.9, calibrated estimate ±20%)
How to run it
Commands for the Q4_K_M file. Both tools download from Hugging Face on first run.
File: Qwen3-30B-A3B-Instruct-2507-Q4_K_M.gguf (18.56 GB) from unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF.
llama-server -hf unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF:Q4_K_M -c 8192 -ngl all -fa on- -ngl all loads every layer on the GPU.
- -fa on enables flash attention (needed for KV cache quantization).
- --cache-type-k q8_0 --cache-type-v q8_0 shrinks the KV cache to 0.43 GB at 8k context.
Terminal:
OLLAMA_CONTEXT_LENGTH=8192 ollama serve
ollama run hf.co/unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF:Q4_K_M- Ollama defaults to 4k context on GPUs under 24 GB. Set OLLAMA_CONTEXT_LENGTH when starting the server, or /set parameter num_ctx inside the chat.
/set parameter num_ctx 8192
Related pages
Guide: Running local LLMs on a Mac (M4, M5): how much unified memory?
Other models on the Apple M4
Qwen3 30B-A3B (2507) on other hardware
Frequently asked questions
Can I run Qwen3 30B-A3B (2507) on the Apple M4?
Q4_K_M fits in unified memory (19.9 of 22.4 GB usable by the GPU): Runs well, est. 12.8 tok/s (10.2–15.4, calibrated estimate ±20%). Q4_K_M needs 19.9 GB at 8k context; the GPU can use 22.4 GB of the unified memory.
How much context can Qwen3 30B-A3B (2507) use on the Apple M4?
At Q4_K_M the verdict stays Runs well up to 16k context with an f16 KV cache (1.6 GB of KV) and up to 32k with a q8_0 KV cache (1.7 GB). The model supports up to 256k.
Which quant should I use, and how fast is it?
Q4_K_M (18.56 GB) is the recommended balance: Runs well, est. 12.8 tok/s (10.2–15.4, calibrated estimate ±20%). It is also the largest tracked file that gets this verdict.
Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.