DeepSeek on NVIDIA Hopper

Can I run DeepSeek-R1 671B-A37B on an H100 SXM 80GB?

Not at Q4_K_M — it needs 380.4 GB against 74.4 GB available. You would need 6 of these cards.

Won't fit

380.4 GB of 74.4 GB · 511%
0396 GB
Weights 379.0 GB
KV cache 0.5 GB
Runtime overhead 0.9 GB
Over the limit 306.0 GB

Short by 306.0 GB. You can run it with 11 of 61 layers on the H100 SXM 80GB and the rest in system RAM, at roughly 2.53 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation2.53tok/s
Prompt processing5613tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of DeepSeek-R1 671B-A37B on a H100 SXM 80GB

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 0.68
INT8 / W8A8 8.50 663.6 GB 665.0 GB Won't fit 1.33
Q8_0 (GGUF) 8.50 663.6 GB 665.0 GB Won't fit 1.33
FP8 (E4M3) 8.00 624.9 GB 626.3 GB Won't fit 1.44
Q6_K 6.56 512.4 GB 513.8 GB Won't fit 1.78
Q5_K_M 5.67 442.9 GB 444.3 GB Won't fit 2.13
Q5_K_S 5.52 431.2 GB 432.6 GB Won't fit 2.19
Q4_K_M 4.85 379.0 GB 380.4 GB Won't fit 2.53
Q4_K_S 4.58 358.0 GB 359.4 GB Won't fit 2.73
Q4_0 4.55 355.6 GB 357.0 GB Won't fit 2.74
AWQ 4-bit 4.25 334.5 GB 335.9 GB Won't fit 2.97
GPTQ 4-bit 4.25 334.5 GB 335.9 GB Won't fit 2.97
MXFP4 4.25 334.5 GB 335.9 GB Won't fit 2.97
IQ4_XS 4.25 332.3 GB 333.7 GB Won't fit 2.99
Q3_K_M 3.91 305.8 GB 307.2 GB Won't fit 3.31
IQ3_M 3.70 289.4 GB 290.8 GB Won't fit 3.56
IQ3_XXS 3.06 239.6 GB 241.0 GB Won't fit 4.55
Q2_K 2.63 206.1 GB 207.5 GB Won't fit 5.63
IQ2_XXS 2.06 161.7 GB 163.1 GB Won't fit 8.22
IQ1_M 1.75 137.5 GB 138.9 GB Won't fit 11.0

Also worth checking