Skip to content

MiniCPM V 4 5 VRAM requirements

How much VRAM MiniCPM V 4 5 needs
4-bit weights: 4.9 GBWith 8K context: 6.9 GBWith 32K context: 10.3 GB8-bit, 8K context: 10.6 GB
Parameters
8.7B
Longest context
40K
Released
Aug 2025
License
Apache 2.0
Strongest at
Text and images

How much VRAM MiniCPM V 4 5 needs

Memory MiniCPM V 4 5 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 40K
16-bit16.2 GB18.2 GB21.6 GB22.7 GB
8-bit8.6 GB10.6 GB14.0 GB15.1 GB
6-bit6.6 GB8.6 GB12.0 GB13.1 GB
5-bit5.8 GB7.7 GB11.1 GB12.2 GB
4-bit4.9 GB6.9 GB10.3 GB11.4 GB
3-bit4.0 GB5.9 GB9.3 GB10.4 GB
2-bit3.0 GB5.0 GB8.4 GB9.5 GB
1.6-bit1.7 GB3.7 GB7.0 GB8.2 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run MiniCPM V 4 5

The best way MiniCPM V 4 5 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40603-bit16K46 tok/s1.1 sPDF: MiniCPM V 4 5 on RTX 4060 report
RTX 50603-bit16K73 tok/s0.9 sPDF: MiniCPM V 4 5 on RTX 5060 report
RTX 30703-bit16K73 tok/s0.8 sPDF: MiniCPM V 4 5 on RTX 3070 report
RTX 3080 10 GB5-bit20K84 tok/s0.6 sPDF: MiniCPM V 4 5 on RTX 3080 10 GB report
RTX 3060 12 GB6-bit24K37 tok/s1.3 sPDF: MiniCPM V 4 5 on RTX 3060 12 GB report
Arc B5806-bit24K32 tok/s0.9 sPDF: MiniCPM V 4 5 on Arc B580 report
RTX 40706-bit24K50 tok/s0.6 sPDF: MiniCPM V 4 5 on RTX 4070 report
RTX 4070 Super6-bit24K50 tok/s0.5 sPDF: MiniCPM V 4 5 on RTX 4070 Super report
RTX 50706-bit24K66 tok/s0.5 sPDF: MiniCPM V 4 5 on RTX 5070 report
RTX 4060 Ti 16 GB8-bit40K23 tok/s0.8 sPDF: MiniCPM V 4 5 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit40K35 tok/s0.7 sPDF: MiniCPM V 4 5 on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit40K30 tok/s0.8 sPDF: MiniCPM V 4 5 on Arc A770 16 GB report
Radeon RX 9070 XT8-bit40K41 tok/s0.5 sPDF: MiniCPM V 4 5 on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit40K52 tok/s0.4 sPDF: MiniCPM V 4 5 on RTX 4070 Ti Super report
RTX 4080 Super8-bit40K56 tok/s0.3 sPDF: MiniCPM V 4 5 on RTX 4080 Super report
RTX 5070 Ti8-bit40K68 tok/s0.4 sPDF: MiniCPM V 4 5 on RTX 5070 Ti report
RTX 50808-bit40K72 tok/s0.3 sPDF: MiniCPM V 4 5 on RTX 5080 report
Radeon RX 7900 XTX8-bit40K60 tok/s0.7 sPDF: MiniCPM V 4 5 on Radeon RX 7900 XTX report
RTX 30908-bit40K71 tok/s0.5 sPDF: MiniCPM V 4 5 on RTX 3090 report
RTX 40908-bit40K76 tok/s0.2 sPDF: MiniCPM V 4 5 on RTX 4090 report
Radeon AI PRO R97008-bit40K41 tok/s0.5 sPDF: MiniCPM V 4 5 on Radeon AI PRO R9700 report
RTX 50908-bit40K126 tok/s0.2 sPDF: MiniCPM V 4 5 on RTX 5090 report
2× RTX 30908-bit40K71 tok/s0.3 sPDF: MiniCPM V 4 5 on 2× RTX 3090 report
2× RTX 40908-bit40K76 tok/s0.1 sPDF: MiniCPM V 4 5 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit40K98 tok/s0.2 sPDF: MiniCPM V 4 5 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit40K72 tok/s0.2 sPDF: MiniCPM V 4 5 on RTX 6000 Ada report
L40S8-bit40K66 tok/s0.3 sPDF: MiniCPM V 4 5 on L40S report
2× RTX 50908-bit40K126 tok/s0.1 sPDF: MiniCPM V 4 5 on 2× RTX 5090 report
4× RTX 30908-bit40K71 tok/s0.1 sPDF: MiniCPM V 4 5 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit40K126 tok/s0.1 sPDF: MiniCPM V 4 5 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit20K14 tok/s11 sPDF: MiniCPM V 4 5 on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit40K11 tok/s11 sPDF: MiniCPM V 4 5 on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit40K11 tok/s11 sPDF: MiniCPM V 4 5 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit40K24 tok/s5.3 sPDF: MiniCPM V 4 5 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit40K24 tok/s5.3 sPDF: MiniCPM V 4 5 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit40K24 tok/s5.3 sPDF: MiniCPM V 4 5 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit40K42 tok/s2.7 sPDF: MiniCPM V 4 5 on M4 Max 128 GB report
M5 Max 128 GB8-bit40K46 tok/s0.8 sPDF: MiniCPM V 4 5 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit40K53 tok/s1.5 sPDF: MiniCPM V 4 5 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit40K53 tok/s1.5 sPDF: MiniCPM V 4 5 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit40K53 tok/s1.5 sPDF: MiniCPM V 4 5 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit40K72 tok/s0.5 sPDF: MiniCPM V 4 5 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit40K17 tok/s1.5 sPDF: MiniCPM V 4 5 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit40K17 tok/s1.5 sPDF: MiniCPM V 4 5 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit40K24 tok/s0.5 sPDF: MiniCPM V 4 5 on RTX Spark report
DGX Spark8-bit40K22 tok/s0.5 sPDF: MiniCPM V 4 5 on DGX Spark report
Dual DGX Spark8-bit40K22 tok/s0.3 sPDF: MiniCPM V 4 5 on Dual DGX Spark report
DGX Station GB3008-bit40K356 tok/s0.1 sPDF: MiniCPM V 4 5 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit40K126 tok/s0.1 sPDF: MiniCPM V 4 5 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit40K126 tok/s0.1 sPDF: MiniCPM V 4 5 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit40K126 tok/s0.1 sPDF: MiniCPM V 4 5 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit40K141 tok/s0.3 sPDF: MiniCPM V 4 5 on A100 80 GB SXM report
H100 SXM8-bit40K211 tok/s0.1 sPDF: MiniCPM V 4 5 on H100 SXM report
4× H200 NVL8-bit40K275 tok/s0.1 sPDF: MiniCPM V 4 5 on 4× H200 NVL report
8× H200 NVL8-bit40K275 tok/s0.1 sPDF: MiniCPM V 4 5 on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf openbmb/MiniCPM-V-4_5-gguf
vLLM
vllm serve openbmb/MiniCPM-V-4_5

Questions

How much VRAM does MiniCPM V 4 5 need?

MiniCPM V 4 5 needs 4.9 GB for its weights at 4-bit, 6.9 GB with an 8K context and 10.3 GB with 32K. At 8-bit, 10.6 GB.

What is the minimum VRAM for MiniCPM V 4 5?

About 5.9 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is MiniCPM V 4 5?

8.7B parameters: 16.2 GB of weights at 16-bit, 4.9 GB at 4-bit.

Can MiniCPM V 4 5 run on an RTX 4090?

Yes. At 8-bit it holds 40K of context and writes about 76 tokens a second.

More models