01 // Representative inference estimates
Single-stream decode · llama.cpp
Aggregated from published community sources. Test setups differ, so use close results as directional evidence.
Sources and normalization →01b // Performance across quantization
vs. nearest competitors
How tok/s scales from FP16 → Q8 → Q4 compared to GPUs in a similar price/VRAM range.
02 // Hardware specs
ArchitectureM4 Max
Process nodeTSMC 3nm
Memory128 GB
Memory bandwidth546 GB/s
FP16 compute34 TFLOPS
INT8 compute68 TOPS
TDP140 W
PCIeUnified
Form factorLaptop
CoolingActive
03 // Model fit
Weight estimate plus 15% planning headroom for runtime buffers and a modest KV cache. Long context can require more.
+ STRENGTHS
- ✓128GB clears our Qwen2.5 72B Q4 planning target
- ✓546 GB/s memory bandwidth · top tier in its class
- ✓Indexed formats: FP16, Q8, Q4, MLX · verify support in your runtime
− TRADE-OFFS
- −Draws 140W under load — plan PSU and thermals accordingly
- −Limited to laptop chassis
- −Mac-only — CUDA tooling won't run
related research
Research behind Apple M4 Max inference tradeoffs
These papers explain the quantization, cache, bandwidth, and runtime constraints that matter before buying this GPU for local AI.
Local AI inference papers
llama.cpp, Apple Silicon, constrained GPUs, offload, and one-box inference.
Open KV cache optimization papers
Cache quantization, compression, reuse, and long-context memory pressure.
Open LLM quantization research
GPTQ, AWQ, GGUF, FP4, NF4, and what low-bit formats mean for VRAM fit.
Open