DeepSeek on Apple M3

Can I run DeepSeek-R1-Distill-Llama 8B on an Mac Studio M3 Ultra 256GB?

Yes. DeepSeek-R1-Distill-Llama 8B at Q4_K_M uses 6.1 GB of the 192.0 GB available on a Mac Studio M3 Ultra 256GB, and runs at about 90.9 tokens per second. You can push the context to 128K.

Runs comfortably

6.1 GB of 192.0 GB · 3%
0192 GB
Weights 4.6 GB
KV cache 1.0 GB
Runtime overhead 0.5 GB

DeepSeek-R1-Distill-Llama 8B at Q4_K_M leaves 185.9 GB spare on a Mac Studio M3 Ultra 256GB. There is room to raise the context length or move up a quantisation level.

Generation90.9tok/s
Prompt processing1491tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of DeepSeek-R1-Distill-Llama 8B 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 15.0 GB 16.5 GB Runs comfortably 128K 30.4
INT8 / W8A8 8.50 7.7 GB 9.2 GB Runs comfortably 128K 57.0
Q8_0 (GGUF) 8.50 7.7 GB 9.2 GB Runs comfortably 128K 57.0
FP8 (E4M3) 8.00 7.5 GB 9.0 GB Runs comfortably 128K 58.6
Q6_K 6.56 6.1 GB 7.7 GB Runs comfortably 128K 70.4
AWQ 4-bit 4.25 5.4 GB 6.9 GB Runs comfortably 128K 78.9
GPTQ 4-bit 4.25 5.4 GB 6.9 GB Runs comfortably 128K 78.9
MXFP4 4.25 5.4 GB 6.9 GB Runs comfortably 128K 78.9
Q5_K_M 5.67 5.3 GB 6.8 GB Runs comfortably 128K 80.3
Q5_K_S 5.52 5.2 GB 6.7 GB Runs comfortably 128K 82.3
Q4_K_M 4.85 4.6 GB 6.1 GB Runs comfortably 128K 90.9
Q4_K_S 4.58 4.4 GB 5.9 GB Runs comfortably 128K 94.9
Q4_0 4.55 4.4 GB 5.9 GB Runs comfortably 128K 95.4
IQ4_XS 4.25 4.1 GB 5.6 GB Runs comfortably 128K 100
Q3_K_M 3.91 3.8 GB 5.4 GB Runs comfortably 128K 107
IQ3_M 3.70 3.7 GB 5.2 GB Runs comfortably 128K 111
IQ3_XXS 3.06 3.2 GB 4.7 GB Runs comfortably 128K 126
Q2_K 2.63 2.8 GB 4.3 GB Runs comfortably 128K 139
IQ2_XXS 2.06 2.3 GB 3.9 GB Runs comfortably 128K 161
IQ1_M 1.75 2.1 GB 3.6 GB Runs comfortably 128K 176

Also worth checking