Qwen on Apple M3

Can I run Qwen2.5 7B on an Mac Studio M3 Ultra 256GB?

Yes. Qwen2.5 7B at Q4_K_M uses 5.3 GB of the 192.0 GB available on a Mac Studio M3 Ultra 256GB, and runs at about 101 tokens per second. You can push the context to 128K.

Runs comfortably

5.3 GB of 192.0 GB · 3%
0192 GB
Weights 4.4 GB
KV cache 0.4 GB
Runtime overhead 0.5 GB

Qwen2.5 7B at Q4_K_M leaves 186.7 GB spare on a Mac Studio M3 Ultra 256GB. There is room to raise the context length or move up a quantisation level.

Generation101tok/s
Prompt processing1571tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen2.5 7B on a Mac Studio M3 Ultra 256GB

Highlighted row is the highest quality that still fits at 8K context.

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 14.2 GB 15.1 GB Runs comfortably 128K 32.7
INT8 / W8A8 8.50 7.3 GB 8.2 GB Runs comfortably 128K 62.3
Q8_0 (GGUF) 8.50 7.3 GB 8.2 GB Runs comfortably 128K 62.3
FP8 (E4M3) 8.00 7.1 GB 8.0 GB Runs comfortably 128K 64.0
Q6_K 6.56 5.8 GB 6.8 GB Runs comfortably 128K 77.4
AWQ 4-bit 4.25 5.3 GB 6.2 GB Runs comfortably 128K 85.1
GPTQ 4-bit 4.25 5.3 GB 6.2 GB Runs comfortably 128K 85.1
MXFP4 4.25 5.3 GB 6.2 GB Runs comfortably 128K 85.1
Q5_K_M 5.67 5.0 GB 6.0 GB Runs comfortably 128K 88.8
Q5_K_S 5.52 4.9 GB 5.8 GB Runs comfortably 128K 91.1
Q4_K_M 4.85 4.4 GB 5.3 GB Runs comfortably 128K 101
Q4_K_S 4.58 4.2 GB 5.1 GB Runs comfortably 128K 106
Q4_0 4.55 4.2 GB 5.1 GB Runs comfortably 128K 106
IQ4_XS 4.25 3.9 GB 4.9 GB Runs comfortably 128K 112
Q3_K_M 3.91 3.7 GB 4.6 GB Runs comfortably 128K 119
IQ3_M 3.70 3.5 GB 4.5 GB Runs comfortably 128K 124
IQ3_XXS 3.06 3.0 GB 4.0 GB Runs comfortably 128K 142
Q2_K 2.63 2.7 GB 3.6 GB Runs comfortably 128K 158
IQ2_XXS 2.06 2.3 GB 3.2 GB Runs comfortably 128K 185
IQ1_M 1.75 2.0 GB 3.0 GB Runs comfortably 128K 203

Also worth checking

Qwen2.5 7B on every GPU
Full hardware tableEverything that runs on a Mac Studio M3 Ultra 256GB
Full model tableWhy the KV cache is the part that bites
Guide