Cohere on NVIDIA Blackwell

Can I run Command R 35B on an RTX 5090?

Not at Q4_K_M — it needs 31.0 GB against 29.3 GB available. Drop to IQ4_XS and it fits, at about 51.1 tokens per second.

Won't fit

31.0 GB of 29.3 GB · 106%
032 GB
Weights 20.1 GB
KV cache 10.0 GB
Runtime overhead 0.9 GB
Over the limit 1.7 GB

Short by 1.7 GB. You can run it with 36 of 40 layers on the RTX 5090 and the rest in system RAM, at roughly 12.7 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation12.7tok/s
Prompt processing1255tok/s
Max context7Ktokens
KV per 1K tokens0GB

Every quantisation of Command R 35B on a RTX 5090

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 0.82
INT8 / W8A8 8.50 33.7 GB 44.6 GB Won't fit 2.21
Q8_0 (GGUF) 8.50 33.7 GB 44.6 GB Won't fit 2.21
FP8 (E4M3) 8.00 32.6 GB 43.5 GB Won't fit 2.39
Q6_K 6.56 26.7 GB 37.6 GB Won't fit 1K 3.81
Q5_K_M 5.67 23.1 GB 34.0 GB Won't fit 4K 5.94
AWQ 4-bit 4.25 23.0 GB 33.9 GB Won't fit 4K 5.95
GPTQ 4-bit 4.25 23.0 GB 33.9 GB Won't fit 4K 5.95
MXFP4 4.25 23.0 GB 33.9 GB Won't fit 4K 5.95
Q5_K_S 5.52 22.5 GB 33.4 GB Won't fit 5K 6.71
Q4_K_M 4.85 20.1 GB 31.0 GB Won't fit 7K 12.7
Q4_K_S 4.58 19.1 GB 30.0 GB Won't fit 7K 20.8
Q4_0 4.55 19.0 GB 29.9 GB Won't fit 8K 20.9
IQ4_XS 4.25 17.9 GB 28.8 GB Fits, but tight 8K 51.1
Q3_K_M 3.91 16.7 GB 27.6 GB Fits, but tight 9K 53.9
IQ3_M 3.70 15.9 GB 26.8 GB Fits, but tight 10K 55.8
IQ3_XXS 3.06 13.7 GB 24.5 GB Runs comfortably 12K 62.5
Q2_K 2.63 12.1 GB 23.0 GB Runs comfortably 13K 68.1
IQ2_XXS 2.06 10.1 GB 21.0 GB Runs comfortably 15K 77.1
IQ1_M 1.75 9.0 GB 19.8 GB Runs comfortably 16K 83.2

Also worth checking