LLM Roofline
Apple · UMA · Backend: mlx

Apple Mac Mini M4 (16 GB) — LLM Sizing & Speed Matrix

Unified LPDDR5X-7500 memory; entry-level 8B Q4/MLX node with near-silent idle draw.

Reference Pricing
$599 Street
Usable Memory
10 GiB (11.5 tuned)
Memory Bandwidth
120 GB/s (72% eff.)
FP16 Compute
8.6 TFLOPS
Roofline Ridge
55.7 FLOP/B
Power (Load / Idle)
45W / 6W

Buy or Rent Apple Mac Mini M4 (16 GB)

Vendor & Cloud Links
Mac Mini M4 on Amazon
Amazon · Direct Hardware
↗
Mac Mini M4 at B&H Photo
B&H Photo · Direct Hardware
↗

All 32 Open-Weight Models on Apple Mac Mini M4 (16 GB) (Q4_K_M @ 32k)

Model Status Weights Total @ 32k Decode TTFT (4k) Max Safe Ctx
Qwen2.5 7B Instruct COMFORTABLE 4.4 GiB 7.2 GiB 13.1 tok/s 13.66s 32k
Gemma 3 4B IT COMFORTABLE 2.5 GiB 4.2 GiB 24.1 tok/s 7.83s 128k
Llama 3.1 8B Instruct TIGHT 4.6 GiB 9.7 GiB 9.3 tok/s 14.61s 46k
Gemma 3 12B IT TUNABLE_UMA 7.0 GiB 10.4 GiB 8.6 tok/s 22.16s 49k
Qwen3 8B TUNABLE_UMA 4.7 GiB 10.3 GiB 8.7 tok/s 15.01s 40k
DeepSeek-R1 671B OOM 386.4 GiB 390.1 GiB OOM — OOM
DeepSeek-V3 671B OOM 386.4 GiB 390.1 GiB OOM — OOM
Llama 3.1 405B Instruct OOM 233.8 GiB 252.3 GiB OOM — OOM
Llama 4 Maverick 400B-A17B OOM 231.5 GiB 238.8 GiB OOM — OOM
Qwen3 235B-A22B OOM 135.4 GiB 142.4 GiB OOM — OOM
Mixtral 8x22B Instruct OOM 81.2 GiB 89.6 GiB OOM — OOM
Command A 111B OOM 63.9 GiB 68.9 GiB OOM — OOM
Llama 4 Scout 109B-A17B OOM 62.8 GiB 70.0 GiB OOM — OOM
Qwen2.5 72B Instruct OOM 41.9 GiB 53.5 GiB OOM — OOM
DeepSeek-R1 Distill Llama 70B OOM 40.7 GiB 52.3 GiB OOM — OOM
Llama 3.3 70B Instruct OOM 40.7 GiB 52.3 GiB OOM — OOM
Llama 3.1 70B Instruct OOM 40.7 GiB 52.3 GiB OOM — OOM
Llama-3.3 Nemotron Super 49B OOM 28.7 GiB 36.5 GiB OOM — OOM
Mixtral 8x7B Instruct OOM 26.9 GiB 32.0 GiB OOM — OOM
Qwen3 32B OOM 18.9 GiB 28.1 GiB OOM — OOM
Qwen2.5 Coder 32B OOM 18.9 GiB 28.1 GiB OOM — OOM
DeepSeek-R1 Distill Qwen 32B OOM 18.9 GiB 28.1 GiB OOM — OOM
QwQ 32B Reasoning OOM 18.9 GiB 28.1 GiB OOM — OOM
GLM-4 32B 0414 OOM 18.6 GiB 21.9 GiB OOM — OOM
Qwen3 30B-A3B OOM 17.6 GiB 21.4 GiB OOM — OOM
Gemma 3 27B IT OOM 15.8 GiB 20.0 GiB OOM — OOM
Mistral Small 3.1 24B OOM 13.8 GiB 20.1 GiB OOM — OOM
Codestral 22B OOM 12.8 GiB 21.2 GiB OOM — OOM
Qwen3 14B OOM 8.5 GiB 14.8 GiB OOM — 11k
Qwen2.5 14B Instruct OOM 8.5 GiB 15.8 GiB OOM — 9k
DeepSeek-R1 Distill Qwen 14B OOM 8.5 GiB 15.8 GiB OOM — 9k
Phi-4 14B OOM 8.5 GiB 12.8 GiB OOM — 9k

Interactive Apple Mac Mini M4 (16 GB) Simulator

7.6B Dense · GQA
120 GB/s · 8.6 TFLOPS
3. Weight Quantization 4.85 bpw
4. KV Cache
32,768 tok
Compute Roofline
COMFORTABLE FIT Memory-Bandwidth Bound
Max Safe Ctx: 32k
Memory Allocation Footprint 7.17 / 10.00 GiB usable
Weights: 4.38 GiB KV Cache: 1.75 GiB Overhead: 1.04 GiB Headroom: 2.83 GiB
Decode Speed
13.1
tok/s (batch=1)
Prefill Speed
301
prompt tok/s
TTFT Latency
13.66s
for 4,096 tok
Active Params
7.6B
of 7.6B total
llama.cpp Server Command 
llama-server -hf Qwen/Qwen2.5-7B-Instruct:Q4_K_M -ngl 99 -fa on -c 32768 --port 8080
Context KV Cache Total VRAM Decode Status
4k 0.22 GiB 5.39 GiB 17.5 tok/s COMFORTABLE
8k 0.44 GiB 5.86 GiB 16.7 tok/s COMFORTABLE
16k 0.88 GiB 6.29 GiB 15.3 tok/s COMFORTABLE
32k 1.75 GiB 7.17 GiB 13.1 tok/s COMFORTABLE

Buy-vs-Cloud-vs-API Monthly TCO Break-Even

USE SERVERLESS API
Own Hardware/mo
$14.16
363.5h active + power
Cloud GPU/mo
$372.09
@$0.89/hr spot/pod
Serverless API/mo
$3.80
API cheaper

Recommended Hardware & Cloud Deployment

Contextual to fit status

Hardware FAQ — Apple Mac Mini M4 (16 GB)

How much usable VRAM / GPU memory does Apple Mac Mini M4 (16 GB) actually provide for LLMs?

Out of the box, Apple Mac Mini M4 (16 GB) safely reserves 10 GiB for GPU inference, expandable up to 11.5 GiB with OS UMA tuning.

What is the largest Q4_K_M model Apple Mac Mini M4 (16 GB) can run at 32k context?

With Q4_K_M weights and FP16 KV cache at 32k context, the largest fitting model in our verified catalog on Apple Mac Mini M4 (16 GB) is Qwen2.5 7B Instruct (7.6B total).