Mistral on NVIDIA Ada

Can I run Mixtral 8x7B on an L40S?

Yes. Mixtral 8x7B at Q4_K_M uses 28.2 GB of the 44.3 GB available on a L40S, and runs at about 55.1 tokens per second. You can push the context to 32K.

Runs comfortably

28.2 GB of 44.3 GB · 64%
044 GB
Weights 26.4 GB
KV cache 1.0 GB
Runtime overhead 0.8 GB

Mixtral 8x7B at Q4_K_M leaves 16.1 GB spare on a L40S. There is room to raise the context length or move up a quantisation level.

Generation55.1tok/s
Prompt processing2947tok/s
Max context32Ktokens
KV per 1K tokens0GB

Every quantisation of Mixtral 8x7B on a L40S

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 87.0 GB 88.8 GB Won't fit — 2.73
INT8 / W8A8 8.50 46.2 GB 48.0 GB Won't fit — 16.6
Q8_0 (GGUF) 8.50 46.2 GB 48.0 GB Won't fit — 16.6
FP8 (E4M3) 8.00 43.5 GB 45.3 GB Won't fit — 26.2
Q6_K 6.56 35.7 GB 37.5 GB Runs comfortably 32K 41.8
Q5_K_M 5.67 30.8 GB 32.6 GB Runs comfortably 32K 47.8
Q5_K_S 5.52 30.0 GB 31.8 GB Runs comfortably 32K 49.0
Q4_K_M 4.85 26.4 GB 28.2 GB Runs comfortably 32K 55.1
Q4_K_S 4.58 24.9 GB 26.7 GB Runs comfortably 32K 58.0
Q4_0 4.55 24.8 GB 26.6 GB Runs comfortably 32K 58.3
AWQ 4-bit 4.25 23.5 GB 25.3 GB Runs comfortably 32K 61.2
GPTQ 4-bit 4.25 23.5 GB 25.3 GB Runs comfortably 32K 61.2
MXFP4 4.25 23.5 GB 25.3 GB Runs comfortably 32K 61.2
IQ4_XS 4.25 23.1 GB 25.0 GB Runs comfortably 32K 61.9
Q3_K_M 3.91 21.3 GB 23.1 GB Runs comfortably 32K 66.6
IQ3_M 3.70 20.2 GB 22.0 GB Runs comfortably 32K 69.9
IQ3_XXS 3.06 16.7 GB 18.5 GB Runs comfortably 32K 82.2
Q2_K 2.63 14.4 GB 16.2 GB Runs comfortably 32K 93.3
IQ2_XXS 2.06 11.3 GB 13.1 GB Runs comfortably 32K 114
IQ1_M 1.75 9.6 GB 11.4 GB Runs comfortably 32K 129

Also worth checking

Mixtral 8x7B on every GPU
Full hardware tableEverything that runs on a L40S
Full model tableWhy the KV cache is the part that bites
Guide