Mistral on NVIDIA Ampere

Can I run Mistral NeMo 12B on an RTX 3080 10GB?

Not at Q4_K_M — it needs 9.1 GB against 8.6 GB available. Drop to IQ4_XS and it fits, at about 68.0 tokens per second.

Won't fit

9.1 GB of 8.6 GB · 106%
09 GB
Weights 7.0 GB
KV cache 1.3 GB
Runtime overhead 0.8 GB
Over the limit 0.5 GB

Short by 0.5 GB. You can run it with 37 of 40 layers on the RTX 3080 10GB and the rest in system RAM, at roughly 34.0 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation34.0tok/s
Prompt processing1029tok/s
Max context5Ktokens
KV per 1K tokens0GB

Every quantisation of Mistral NeMo 12B on a RTX 3080 10GB

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 22.8 GB 24.9 GB Won't fit 2.28
INT8 / W8A8 8.50 11.8 GB 13.9 GB Won't fit 6.46
Q8_0 (GGUF) 8.50 11.8 GB 13.9 GB Won't fit 6.46
FP8 (E4M3) 8.00 11.4 GB 13.5 GB Won't fit 6.68
Q6_K 6.56 9.4 GB 11.4 GB Won't fit 10.5
Q5_K_M 5.67 8.1 GB 10.2 GB Won't fit 17.1
AWQ 4-bit 4.25 7.9 GB 10.0 GB Won't fit 17.5
GPTQ 4-bit 4.25 7.9 GB 10.0 GB Won't fit 17.5
MXFP4 4.25 7.9 GB 10.0 GB Won't fit 17.5
Q5_K_S 5.52 7.9 GB 10.0 GB Won't fit 19.2
Q4_K_M 4.85 7.0 GB 9.1 GB Won't fit 5K 34.0
Q4_K_S 4.58 6.7 GB 8.8 GB Won't fit 7K 50.6
Q4_0 4.55 6.6 GB 8.7 GB Won't fit 7K 50.8
IQ4_XS 4.25 6.3 GB 8.3 GB Fits, but tight 10K 68.0
Q3_K_M 3.91 5.8 GB 7.9 GB Fits, but tight 12K 72.5
IQ3_M 3.70 5.6 GB 7.6 GB Runs comfortably 14K 75.5
IQ3_XXS 3.06 4.7 GB 6.8 GB Runs comfortably 19K 86.7
Q2_K 2.63 4.2 GB 6.3 GB Runs comfortably 23K 96.3
IQ2_XXS 2.06 3.5 GB 5.6 GB Runs comfortably 27K 113
IQ1_M 1.75 3.1 GB 5.2 GB Runs comfortably 30K 125

Also worth checking