Cohere on NVIDIA Ada

Can I run Command R 35B on an RTX 4070 Super?

Not at Q4_K_M — it needs 31.0 GB against 10.5 GB available. You would need 3 of these cards.

Won't fit

31.0 GB of 10.5 GB · 295%
032 GB
Weights 20.1 GB
KV cache 10.0 GB
Runtime overhead 0.9 GB
Over the limit 20.5 GB

Short by 20.5 GB. You can run it with 0 of 40 layers on the RTX 4070 Super and the rest in system RAM, at roughly 1.53 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation1.53tok/s
Prompt processing426tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of Command R 35B on a RTX 4070 Super

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 65.2 GB 76.0 GB Won't fit 0.55
INT8 / W8A8 8.50 33.7 GB 44.6 GB Won't fit 0.99
Q8_0 (GGUF) 8.50 33.7 GB 44.6 GB Won't fit 0.99
FP8 (E4M3) 8.00 32.6 GB 43.5 GB Won't fit 1.02
Q6_K 6.56 26.7 GB 37.6 GB Won't fit 1.21
Q5_K_M 5.67 23.1 GB 34.0 GB Won't fit 1.36
AWQ 4-bit 4.25 23.0 GB 33.9 GB Won't fit 1.36
GPTQ 4-bit 4.25 23.0 GB 33.9 GB Won't fit 1.36
MXFP4 4.25 23.0 GB 33.9 GB Won't fit 1.36
Q5_K_S 5.52 22.5 GB 33.4 GB Won't fit 1.39
Q4_K_M 4.85 20.1 GB 31.0 GB Won't fit 1.53
Q4_K_S 4.58 19.1 GB 30.0 GB Won't fit 1.59
Q4_0 4.55 19.0 GB 29.9 GB Won't fit 1.59
IQ4_XS 4.25 17.9 GB 28.8 GB Won't fit 1.67
Q3_K_M 3.91 16.7 GB 27.6 GB Won't fit 1.76
IQ3_M 3.70 15.9 GB 26.8 GB Won't fit 1.83
IQ3_XXS 3.06 13.7 GB 24.5 GB Won't fit 2.05
Q2_K 2.63 12.1 GB 23.0 GB Won't fit 2.24
IQ2_XXS 2.06 10.1 GB 21.0 GB Won't fit 2.54
IQ1_M 1.75 9.0 GB 19.8 GB Won't fit 512 2.74

Also worth checking