llamaperf

M1 Ultra 128GB vs M3 Ultra 256GB

For running local LLMs · 4 reports across 4 models

Side A
M1 Ultra 128GB
Vendor
apple
VRAM
128GB
Memory
Unified
Side B
M3 Ultra 256GB
Vendor
apple
VRAM
256GB
Memory
Unified

Highest reported generation speed by model family

These maxima come from different community setups. Model size, quantization, GPU count, offloading and concurrency may differ. They do not establish which GPU is faster for the same workload. Check the comparison conditions →

ModelM1 Ultra 128GBM3 Ultra 256GB
DeepSeek V4up to 284B16.0n=1
GLM-5.3up to 321B37.4n=1
Minimax M317.3n=1
Qwen3.8up to 2400B31.2n=1

More comparisons