Model Anatomy See what's inside open-weight AI models

Can I run MiniCPM-V-4.7-35B-A3B-gguf on an RTX 4090?

Yes, just

Its Q4_K_M file (21.2 GB) fits in the 21.6 GB an RTX 4090 can use, with room for ≈ 18K tokens of context.

Full analysis of MiniCPM-V-4.7-35B-A3B-gguf Everything that runs on RTX 4090 Get the Q4_K_M file on Hugging Face
ollama run hf.co/openbmb/MiniCPM-V-4.7-35B-A3B-gguf:Q4_K_M

Files and context on RTX 4090

Each file of MiniCPM-V-4.7-35B-A3B-gguf (and of GGUF versions the site has analyzed), and how much context fits next to it, with the context cache in 16-bit or 8-bit.

FileSizeOn RTX 4090 Context (16-bit)Context (8-bit)
F1669.4 GBDoesn't fit——
MiniCPM-V-4_7-35B-A3B-F16-thinking69.4 GBDoesn't fit——
Q8_036.9 GBDoesn't fit——
MiniCPM-V-4_7-35B-A3B-Q8_0-thinking36.9 GBDoesn't fit——
Q6_K28.5 GBDoesn't fit——
MiniCPM-V-4_7-35B-A3B-Q6_K-thinking28.5 GBDoesn't fit——
Q5_126.1 GBDoesn't fit——
MiniCPM-V-4_7-35B-A3B-Q5_1-thinking26.1 GBDoesn't fit——
Q5_K_M24.7 GBDoesn't fit——
MiniCPM-V-4_7-35B-A3B-Q5_K_M-thinking24.7 GBDoesn't fit——
Q5_024.0 GBDoesn't fit——
Q5_K_S24.0 GBDoesn't fit——
MiniCPM-V-4_7-35B-A3B-Q5_0-thinking24.0 GBDoesn't fit——
MiniCPM-V-4_7-35B-A3B-Q5_K_S-thinking24.0 GBDoesn't fit——
Q4_121.8 GBDoesn't fit——
MiniCPM-V-4_7-35B-A3B-Q4_1-thinking21.8 GBDoesn't fit——
Q4_K_M21.2 GBTight18K35K
MiniCPM-V-4_7-35B-A3B-Q4_K_M-thinking21.2 GBTight18K35K
Q4_K_S19.9 GBTight78K157K
MiniCPM-V-4_7-35B-A3B-Q4_K_S-thinking19.9 GBTight78K157K
IQ4_NL19.8 GBTight84K168K
MiniCPM-V-4_7-35B-A3B-IQ4_NL-thinking19.8 GBTight84K168K
Q4_019.7 GBTight87K174K
MiniCPM-V-4_7-35B-A3B-Q4_0-thinking19.7 GBTight87K174K
IQ4_XS18.7 GBTight134K256K
MiniCPM-V-4_7-35B-A3B-IQ4_XS-thinking18.7 GBTight134K256K
MiniCPM-V-4_7-35B-A3B-mix_q3gud_im15.5 GBFits256K256K
MiniCPM-V-4_7-35B-A3B-mix_q3gud_im-thinking15.5 GBFits256K256K
IQ3_M15.4 GBFits256K256K
MiniCPM-V-4_7-35B-A3B-IQ3_M-thinking15.4 GBFits256K256K
IQ3_XS14.5 GBFits256K256K
MiniCPM-V-4_7-35B-A3B-IQ3_XS-thinking14.5 GBFits256K256K

MiniCPM-V-4.7-35B-A3B-gguf on other hardware

Popular models that run on RTX 4090

Assumes 90% of the card's memory is usable and at least 4K tokens of context for one conversation, with the context cache in 16-bit unless noted. Fits leaves at least 20% of that memory for context; Tight leaves less.