Mistral on NVIDIA Ada

Can I run Mistral NeMo 12B on an L40S?

Yes. Mistral NeMo 12B at Q4_K_M uses 9.1 GB of the 44.3 GB available on a L40S, and runs at about 66.7 tokens per second. You can push the context to 128K.

Runs comfortably

9.1 GB of 44.3 GB · 21%
044 GB
Weights 7.0 GB
KV cache 1.3 GB
Runtime overhead 0.8 GB

Mistral NeMo 12B at Q4_K_M leaves 35.2 GB spare on a L40S. There is room to raise the context length or move up a quantisation level.

Generation66.7tok/s
Prompt processing3103tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Mistral NeMo 12B on a L40S

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 Runs comfortably 128K 22.0
INT8 / W8A8 8.50 11.8 GB 13.9 GB Runs comfortably 128K 41.2
Q8_0 (GGUF) 8.50 11.8 GB 13.9 GB Runs comfortably 128K 41.2
FP8 (E4M3) 8.00 11.4 GB 13.5 GB Runs comfortably 128K 42.6
Q6_K 6.56 9.4 GB 11.4 GB Runs comfortably 128K 51.3
Q5_K_M 5.67 8.1 GB 10.2 GB Runs comfortably 128K 58.6
AWQ 4-bit 4.25 7.9 GB 10.0 GB Runs comfortably 128K 59.9
GPTQ 4-bit 4.25 7.9 GB 10.0 GB Runs comfortably 128K 59.9
MXFP4 4.25 7.9 GB 10.0 GB Runs comfortably 128K 59.9
Q5_K_S 5.52 7.9 GB 10.0 GB Runs comfortably 128K 60.1
Q4_K_M 4.85 7.0 GB 9.1 GB Runs comfortably 128K 66.7
Q4_K_S 4.58 6.7 GB 8.8 GB Runs comfortably 128K 69.8
Q4_0 4.55 6.6 GB 8.7 GB Runs comfortably 128K 70.1
IQ4_XS 4.25 6.3 GB 8.3 GB Runs comfortably 128K 74.0
Q3_K_M 3.91 5.8 GB 7.9 GB Runs comfortably 128K 78.8
IQ3_M 3.70 5.6 GB 7.6 GB Runs comfortably 128K 82.1
IQ3_XXS 3.06 4.7 GB 6.8 GB Runs comfortably 128K 94.3
Q2_K 2.63 4.2 GB 6.3 GB Runs comfortably 128K 105
IQ2_XXS 2.06 3.5 GB 5.6 GB Runs comfortably 128K 123
IQ1_M 1.75 3.1 GB 5.2 GB Runs comfortably 128K 135

Also worth checking

Mistral NeMo 12B on every GPU
Full hardware tableEverything that runs on a L40S
Full model tableWhy the KV cache is the part that bites
Guide