Llama on Apple M3

Can I run Llama 3.2 3B on an Mac Studio M3 Ultra 256GB?

Yes. Llama 3.2 3B at Q4_K_M uses 3.2 GB of the 192.0 GB available on a Mac Studio M3 Ultra 256GB, and runs at about 201 tokens per second. You can push the context to 128K.

Runs comfortably

3.2 GB of 192.0 GB · 2%
0192 GB
Weights 1.8 GB
KV cache 0.9 GB
Runtime overhead 0.5 GB

Llama 3.2 3B at Q4_K_M leaves 188.8 GB spare on a Mac Studio M3 Ultra 256GB. There is room to raise the context length or move up a quantisation level.

Generation201tok/s
Prompt processing3729tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Llama 3.2 3B 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 6.0 GB 7.4 GB Runs comfortably 128K 72.8
INT8 / W8A8 8.50 3.1 GB 4.5 GB Runs comfortably 128K 131
Q8_0 (GGUF) 8.50 3.1 GB 4.5 GB Runs comfortably 128K 131
FP8 (E4M3) 8.00 3.0 GB 4.4 GB Runs comfortably 128K 135
Q6_K 6.56 2.5 GB 3.8 GB Runs comfortably 128K 159
AWQ 4-bit 4.25 2.1 GB 3.5 GB Runs comfortably 128K 179
GPTQ 4-bit 4.25 2.1 GB 3.5 GB Runs comfortably 128K 179
MXFP4 4.25 2.1 GB 3.5 GB Runs comfortably 128K 179
Q5_K_M 5.67 2.1 GB 3.5 GB Runs comfortably 128K 180
Q5_K_S 5.52 2.1 GB 3.4 GB Runs comfortably 128K 183
Q4_K_M 4.85 1.8 GB 3.2 GB Runs comfortably 128K 201
Q4_K_S 4.58 1.8 GB 3.1 GB Runs comfortably 128K 208
Q4_0 4.55 1.7 GB 3.1 GB Runs comfortably 128K 209
IQ4_XS 4.25 1.6 GB 3.0 GB Runs comfortably 128K 219
Q3_K_M 3.91 1.5 GB 2.9 GB Runs comfortably 128K 231
IQ3_M 3.70 1.5 GB 2.8 GB Runs comfortably 128K 239
IQ3_XXS 3.06 1.3 GB 2.6 GB Runs comfortably 128K 267
Q2_K 2.63 1.1 GB 2.5 GB Runs comfortably 128K 290
IQ2_XXS 2.06 0.9 GB 2.3 GB Runs comfortably 128K 328
IQ1_M 1.75 0.8 GB 2.2 GB Runs comfortably 128K 352

Also worth checking

Llama 3.2 3B 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