DeepSeek on NVIDIA Ada

Can I run DeepSeek-R1 671B-A37B on an L40S?

Not at Q4_K_M — it needs 380.4 GB against 44.3 GB available. No quantisation of this model fits on one card.

Won't fit

380.4 GB of 44.3 GB · 858%
0396 GB
Weights 379.0 GB
KV cache 0.5 GB
Runtime overhead 0.9 GB
Over the limit 336.1 GB

Short by 336.1 GB. You can run it with 6 of 61 layers on the L40S and the rest in system RAM, at roughly 1.97 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation1.97tok/s
Prompt processing1027tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of DeepSeek-R1 671B-A37B on a L40S

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 1249.8 GB 1251.2 GB Won't fit 0.57
INT8 / W8A8 8.50 663.6 GB 665.0 GB Won't fit 1.08
Q8_0 (GGUF) 8.50 663.6 GB 665.0 GB Won't fit 1.08
FP8 (E4M3) 8.00 624.9 GB 626.3 GB Won't fit 1.17
Q6_K 6.56 512.4 GB 513.8 GB Won't fit 1.44
Q5_K_M 5.67 442.9 GB 444.3 GB Won't fit 1.66
Q5_K_S 5.52 431.2 GB 432.6 GB Won't fit 1.73
Q4_K_M 4.85 379.0 GB 380.4 GB Won't fit 1.97
Q4_K_S 4.58 358.0 GB 359.4 GB Won't fit 2.11
Q4_0 4.55 355.6 GB 357.0 GB Won't fit 2.13
AWQ 4-bit 4.25 334.5 GB 335.9 GB Won't fit 2.26
GPTQ 4-bit 4.25 334.5 GB 335.9 GB Won't fit 2.26
MXFP4 4.25 334.5 GB 335.9 GB Won't fit 2.26
IQ4_XS 4.25 332.3 GB 333.7 GB Won't fit 2.27
Q3_K_M 3.91 305.8 GB 307.2 GB Won't fit 2.51
IQ3_M 3.70 289.4 GB 290.8 GB Won't fit 2.69
IQ3_XXS 3.06 239.6 GB 241.0 GB Won't fit 3.28
Q2_K 2.63 206.1 GB 207.5 GB Won't fit 3.93
IQ2_XXS 2.06 161.7 GB 163.1 GB Won't fit 5.33
IQ1_M 1.75 137.5 GB 138.9 GB Won't fit 6.58

Also worth checking