Home › GPUs & Macs › Apple M1 Pro (16-core GPU) vs Apple M2 Pro (16-core GPU)

Apple M1 Pro (16-core GPU) vs Apple M2 Pro (16-core GPU) for local LLMs

Across 6 tests measured on both, the Apple M2 Pro (16-core GPU) is faster in 5 and generates a median 1.07× the tokens/s of the Apple M1 Pro (16-core GPU).

On Llama 2 7B (F16, llama-bench) the Apple M1 Pro (16-core GPU) generates 12.8 tokens/s and the Apple M2 Pro (16-core GPU) 12.5 tokens/s (measured).

Same model, same quant, same test

ModelQuant / testM1 Pro (16-core GPU)
tok/s
M2 Pro (16-core GPU)
tok/s
FasterSources and run details
Llama 2 7BF16
llama-bench
12.8
Metal
12.5
Metal
evenllama.cpp #4167 M1 Pro (16-core GPU) · llama.cpp #4167 M2 Pro (16-core GPU)
Llama 2 7BQ4_0
llama-bench
36.4
Metal
37.9
Metal
M2 Pro (16-core GPU) +4%llama.cpp #4167 M1 Pro (16-core GPU) · llama.cpp #4167 M2 Pro (16-core GPU)
Llama 2 7BQ8_0
llama-bench
22.3
Metal
22.7
Metal
evenllama.cpp #4167 M1 Pro (16-core GPU) · llama.cpp #4167 M2 Pro (16-core GPU)
Llama 3.1 8B InstructQ4_K_M
LocalScore
21.8
llamafile
24.2
llamafile
M2 Pro (16-core GPU) +11%LocalScore M1 Pro (16-core GPU) · LocalScore M2 Pro (16-core GPU)
Llama 3.2 1B InstructQ4_K_M
LocalScore
77.8
llamafile
91.2
llamafile
M2 Pro (16-core GPU) +17%LocalScore M1 Pro (16-core GPU) · LocalScore M2 Pro (16-core GPU)
Qwen2.5 14B InstructQ4_K_M
LocalScore
11.8
llamafile
13.3
llamafile
M2 Pro (16-core GPU) +14%LocalScore M1 Pro (16-core GPU) · LocalScore M2 Pro (16-core GPU)

Generation speed in tokens/s, best measured backend for each chip (shown under the number). Backends can differ between the two chips; check them before drawing conclusions. Measured results only, nothing estimated.

Specs

Apple M1 Pro (16-core GPU)Apple M2 Pro (16-core GPU)
MemoryUnified memoryUnified memory
Memory bandwidth200 GB/s200 GB/s

All results for the Apple M1 Pro (16-core GPU) · All results for the Apple M2 Pro (16-core GPU) · More comparisons