Mistral on Apple M3

Can I run Mistral 7B v0.3 on an Mac Studio M3 Ultra 256GB?

Yes. Mistral 7B v0.3 at Q4_K_M uses 5.6 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 32K.

Runs comfortably

5.6 GB of 192.0 GB · 3%
0192 GB
Weights 4.1 GB
KV cache 1.0 GB
Runtime overhead 0.5 GB

Mistral 7B v0.3 at Q4_K_M leaves 186.4 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 processing1651tok/s
Max context32Ktokens
KV per 1K tokens0GB

Every quantisation of Mistral 7B v0.3 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 13.5 GB 15.0 GB Runs comfortably 32K 33.6
INT8 / W8A8 8.50 7.1 GB 8.6 GB Runs comfortably 32K 61.4
Q8_0 (GGUF) 8.50 7.1 GB 8.6 GB Runs comfortably 32K 61.4
FP8 (E4M3) 8.00 6.8 GB 8.3 GB Runs comfortably 32K 64.4
Q6_K 6.56 5.5 GB 7.1 GB Runs comfortably 32K 77.2
Q5_K_M 5.67 4.8 GB 6.3 GB Runs comfortably 32K 88.0
Q5_K_S 5.52 4.7 GB 6.2 GB Runs comfortably 32K 90.1
Q4_K_M 4.85 4.1 GB 5.6 GB Runs comfortably 32K 101
AWQ 4-bit 4.25 4.0 GB 5.5 GB Runs comfortably 32K 104
GPTQ 4-bit 4.25 4.0 GB 5.5 GB Runs comfortably 32K 104
MXFP4 4.25 4.0 GB 5.5 GB Runs comfortably 32K 104
Q4_K_S 4.58 3.9 GB 5.4 GB Runs comfortably 32K 106
Q4_0 4.55 3.9 GB 5.4 GB Runs comfortably 32K 106
IQ4_XS 4.25 3.6 GB 5.1 GB Runs comfortably 32K 112
Q3_K_M 3.91 3.3 GB 4.9 GB Runs comfortably 32K 120
IQ3_M 3.70 3.2 GB 4.7 GB Runs comfortably 32K 126
IQ3_XXS 3.06 2.7 GB 4.2 GB Runs comfortably 32K 146
Q2_K 2.63 2.3 GB 3.8 GB Runs comfortably 32K 163
IQ2_XXS 2.06 1.8 GB 3.4 GB Runs comfortably 32K 194
IQ1_M 1.75 1.6 GB 3.1 GB Runs comfortably 32K 217

Also worth checking

Mistral 7B v0.3 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