Apple Silicon
$2,499
Chip
Apple M4 Pro
Unified memory
48 GB
Llama 3.1 8B tok/s
42
Max model size
~40 GB
Best-case est. time to first token — 16k-token prompt (Llama 3.1 8B)
~53.4s
Scaled from a third-party llama.cpp short-prompt prefill measurement (Llama 2 7B Q4_0, 364.06 tok/s prefill — source).