DeepSeek on NVIDIA Ada

Can I run DeepSeek-R1-Distill-Llama 8B on an L40S?

Yes. DeepSeek-R1-Distill-Llama 8B at Q4_K_M uses 6.4 GB of the 44.3 GB available on a L40S, and runs at about 99.4 tokens per second. You can push the context to 128K.

Runs comfortably

6.4 GB of 44.3 GB · 15%
044 GB
Weights 4.6 GB
KV cache 1.0 GB
Runtime overhead 0.8 GB

DeepSeek-R1-Distill-Llama 8B at Q4_K_M leaves 37.9 GB spare on a L40S. There is room to raise the context length or move up a quantisation level.

Generation99.4tok/s
Prompt processing4733tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of DeepSeek-R1-Distill-Llama 8B on a L40S

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 15.0 GB 16.8 GB Runs comfortably 128K 33.3
INT8 / W8A8 8.50 7.7 GB 9.5 GB Runs comfortably 128K 62.4
Q8_0 (GGUF) 8.50 7.7 GB 9.5 GB Runs comfortably 128K 62.4
FP8 (E4M3) 8.00 7.5 GB 9.3 GB Runs comfortably 128K 64.1
Q6_K 6.56 6.1 GB 8.0 GB Runs comfortably 128K 77.0
AWQ 4-bit 4.25 5.4 GB 7.2 GB Runs comfortably 128K 86.2
GPTQ 4-bit 4.25 5.4 GB 7.2 GB Runs comfortably 128K 86.2
MXFP4 4.25 5.4 GB 7.2 GB Runs comfortably 128K 86.2
Q5_K_M 5.67 5.3 GB 7.1 GB Runs comfortably 128K 87.8
Q5_K_S 5.52 5.2 GB 7.0 GB Runs comfortably 128K 90.0
Q4_K_M 4.85 4.6 GB 6.4 GB Runs comfortably 128K 99.4
Q4_K_S 4.58 4.4 GB 6.2 GB Runs comfortably 128K 104
Q4_0 4.55 4.4 GB 6.2 GB Runs comfortably 128K 104
IQ4_XS 4.25 4.1 GB 5.9 GB Runs comfortably 128K 110
Q3_K_M 3.91 3.8 GB 5.7 GB Runs comfortably 128K 116
IQ3_M 3.70 3.7 GB 5.5 GB Runs comfortably 128K 121
IQ3_XXS 3.06 3.2 GB 5.0 GB Runs comfortably 128K 138
Q2_K 2.63 2.8 GB 4.6 GB Runs comfortably 128K 152
IQ2_XXS 2.06 2.3 GB 4.2 GB Runs comfortably 128K 176
IQ1_M 1.75 2.1 GB 3.9 GB Runs comfortably 128K 192

Also worth checking

DeepSeek-R1-Distill-Llama 8B on every GPU
Full hardware tableEverything that runs on a L40S
Full model tableWhy the KV cache is the part that bites
Guide