Home › GPUs & Macs › Apple M1 Max (24-core GPU) vs Apple M1 Max (32-core GPU)

Apple M1 Max (24-core GPU) vs Apple M1 Max (32-core GPU) for local LLMs

Across 6 tests measured on both, the Apple M1 Max (32-core GPU) is faster in 6 and generates a median 1.12× the tokens/s of the Apple M1 Max (24-core GPU).

On Llama 2 7B (F16, llama-bench) the Apple M1 Max (24-core GPU) generates 22.6 tokens/s and the Apple M1 Max (32-core GPU) 23.0 tokens/s (measured).

Same model, same quant, same test

ModelQuant / testM1 Max (24-core GPU)
tok/s
M1 Max (32-core GPU)
tok/s
FasterSources and run details
Llama 2 7BF16
llama-bench
22.6
Metal
23.0
Metal
evenllama.cpp #4167 M1 Max (24-core GPU) · llama.cpp #4167 M1 Max (32-core GPU)
Llama 2 7BQ4_0
llama-bench
54.6
Metal
61.2
Metal
M1 Max (32-core GPU) +12%llama.cpp #4167 M1 Max (24-core GPU) · llama.cpp #4167 M1 Max (32-core GPU)
Llama 2 7BQ8_0
llama-bench
37.8
Metal
40.2
Metal
M1 Max (32-core GPU) +6%llama.cpp #4167 M1 Max (24-core GPU) · llama.cpp #4167 M1 Max (32-core GPU)
Llama 3.1 8B InstructQ4_K_M
LocalScore
32.1
llamafile
36.2
llamafile
M1 Max (32-core GPU) +13%LocalScore M1 Max (24-core GPU) · LocalScore M1 Max (32-core GPU)
Llama 3.2 1B InstructQ4_K_M
LocalScore
99.8
llamafile
121
llamafile
M1 Max (32-core GPU) +22%LocalScore M1 Max (24-core GPU) · LocalScore M1 Max (32-core GPU)
Qwen2.5 14B InstructQ4_K_M
LocalScore
16.7
llamafile
19.8
llamafile
M1 Max (32-core GPU) +19%LocalScore M1 Max (24-core GPU) · LocalScore M1 Max (32-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 Max (24-core GPU)Apple M1 Max (32-core GPU)
MemoryUnified memoryUnified memory
Memory bandwidth400 GB/s400 GB/s

All results for the Apple M1 Max (24-core GPU) · All results for the Apple M1 Max (32-core GPU) · More comparisons