Mistral on Apple M4

Can I run Mixtral 8x7B on an Mac Studio M4 Max 128GB?

Yes. Mixtral 8x7B at Q4_K_M uses 27.9 GB of the 96.0 GB available on a Mac Studio M4 Max 128GB, and runs at about 37.8 tokens per second. You can push the context to 32K.

Runs comfortably

27.9 GB of 96.0 GB · 29%
096 GB
Weights 26.4 GB
KV cache 1.0 GB
Runtime overhead 0.5 GB

Mixtral 8x7B at Q4_K_M leaves 68.1 GB spare on a Mac Studio M4 Max 128GB. There is room to raise the context length or move up a quantisation level.

Generation37.8tok/s
Prompt processing553tok/s
Max context32Ktokens
KV per 1K tokens0GB

Every quantisation of Mixtral 8x7B on a Mac Studio M4 Max 128GB

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 87.0 GB 88.5 GB Fits, but tight 32K 12.2
INT8 / W8A8 8.50 46.2 GB 47.7 GB Runs comfortably 32K 22.5
Q8_0 (GGUF) 8.50 46.2 GB 47.7 GB Runs comfortably 32K 22.5
FP8 (E4M3) 8.00 43.5 GB 45.0 GB Runs comfortably 32K 23.8
Q6_K 6.56 35.7 GB 37.2 GB Runs comfortably 32K 28.6
Q5_K_M 5.67 30.8 GB 32.3 GB Runs comfortably 32K 32.8
Q5_K_S 5.52 30.0 GB 31.5 GB Runs comfortably 32K 33.6
Q4_K_M 4.85 26.4 GB 27.9 GB Runs comfortably 32K 37.8
Q4_K_S 4.58 24.9 GB 26.4 GB Runs comfortably 32K 39.8
Q4_0 4.55 24.8 GB 26.3 GB Runs comfortably 32K 40.0
AWQ 4-bit 4.25 23.5 GB 25.0 GB Runs comfortably 32K 42.0
GPTQ 4-bit 4.25 23.5 GB 25.0 GB Runs comfortably 32K 42.0
MXFP4 4.25 23.5 GB 25.0 GB Runs comfortably 32K 42.0
IQ4_XS 4.25 23.1 GB 24.7 GB Runs comfortably 32K 42.6
Q3_K_M 3.91 21.3 GB 22.8 GB Runs comfortably 32K 45.8
IQ3_M 3.70 20.2 GB 21.7 GB Runs comfortably 32K 48.1
IQ3_XXS 3.06 16.7 GB 18.2 GB Runs comfortably 32K 56.8
Q2_K 2.63 14.4 GB 15.9 GB Runs comfortably 32K 64.6
IQ2_XXS 2.06 11.3 GB 12.8 GB Runs comfortably 32K 78.9
IQ1_M 1.75 9.6 GB 11.1 GB Runs comfortably 32K 89.8

Also worth checking