DeepSeek on NVIDIA Ada

Can I run DeepSeek-R1-Distill-Qwen 32B on an RTX 4070?

Not at Q4_K_M — it needs 21.5 GB against 10.5 GB available. You would need 3 of these cards.

Won't fit

21.5 GB of 10.5 GB · 205%
022 GB
Weights 18.6 GB
KV cache 2.0 GB
Runtime overhead 0.8 GB
Over the limit 11.0 GB

Short by 11.0 GB. You can run it with 26 of 64 layers on the RTX 4070 and the rest in system RAM, at roughly 3.02 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation3.02tok/s
Prompt processing372tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of DeepSeek-R1-Distill-Qwen 32B on a RTX 4070

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 61.0 GB 63.9 GB Won't fit — 0.69
INT8 / W8A8 8.50 32.1 GB 34.9 GB Won't fit — 1.45
Q8_0 (GGUF) 8.50 32.1 GB 34.9 GB Won't fit — 1.45
FP8 (E4M3) 8.00 30.5 GB 33.3 GB Won't fit — 1.55
Q6_K 6.56 25.0 GB 27.9 GB Won't fit — 1.98
Q5_K_M 5.67 21.6 GB 24.5 GB Won't fit — 2.42
Q5_K_S 5.52 21.1 GB 23.9 GB Won't fit — 2.53
Q4_K_M 4.85 18.6 GB 21.5 GB Won't fit — 3.02
AWQ 4-bit 4.25 18.3 GB 21.2 GB Won't fit — 3.06
GPTQ 4-bit 4.25 18.3 GB 21.2 GB Won't fit — 3.06
MXFP4 4.25 18.3 GB 21.2 GB Won't fit — 3.06
Q4_K_S 4.58 17.6 GB 20.5 GB Won't fit — 3.25
Q4_0 4.55 17.5 GB 20.4 GB Won't fit — 3.27
IQ4_XS 4.25 16.4 GB 19.3 GB Won't fit — 3.63
Q3_K_M 3.91 15.2 GB 18.0 GB Won't fit — 4.19
IQ3_M 3.70 14.4 GB 17.3 GB Won't fit — 4.50
IQ3_XXS 3.06 12.1 GB 15.0 GB Won't fit — 6.41
Q2_K 2.63 10.6 GB 13.4 GB Won't fit — 8.87
IQ2_XXS 2.06 8.5 GB 11.3 GB Won't fit 5K 18.0
IQ1_M 1.75 7.4 GB 10.2 GB Fits, but tight 9K 35.7

Also worth checking

DeepSeek-R1-Distill-Qwen 32B on every GPU
Full hardware tableEverything that runs on a RTX 4070
Full model tableWhy the KV cache is the part that bites
Guide