Skip to content

MiniCPM V 4 VRAM requirements

How much VRAM MiniCPM V 4 needs
4-bit weights: 2.3 GBWith 8K context: 3.2 GBWith 32K context: 4.0 GB8-bit, 8K context: 4.9 GB
Parameters
4.1B
Longest context
32K
Released
Jul 2025
License
Apache 2.0
Strongest at
Text and images

How much VRAM MiniCPM V 4 needs

Memory MiniCPM V 4 needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit7.6 GB8.5 GB9.2 GB
8-bit4.0 GB4.9 GB5.7 GB
6-bit3.1 GB4.0 GB4.8 GB
5-bit2.7 GB3.6 GB4.4 GB
4-bit2.3 GB3.2 GB4.0 GB
3-bit1.8 GB2.8 GB3.5 GB
2-bit1.4 GB2.3 GB3.1 GB
1.6-bit798 MB1.7 GB2.4 GB

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

GPUs and Macs that run MiniCPM V 4

The best way MiniCPM V 4 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit32K44 tok/s0.5 sPDF: MiniCPM V 4 on RTX 4060 report
RTX 50608-bit32K71 tok/s0.4 sPDF: MiniCPM V 4 on RTX 5060 report
RTX 30708-bit32K71 tok/s0.4 sPDF: MiniCPM V 4 on RTX 3070 report
RTX 3080 10 GB8-bit32K115 tok/s0.3 sPDF: MiniCPM V 4 on RTX 3080 10 GB report
RTX 3060 12 GB8-bit32K58 tok/s0.7 sPDF: MiniCPM V 4 on RTX 3060 12 GB report
Arc B5808-bit32K50 tok/s0.5 sPDF: MiniCPM V 4 on Arc B580 report
RTX 40708-bit32K79 tok/s0.3 sPDF: MiniCPM V 4 on RTX 4070 report
RTX 4070 Super8-bit32K79 tok/s0.2 sPDF: MiniCPM V 4 on RTX 4070 Super report
RTX 50708-bit32K103 tok/s0.3 sPDF: MiniCPM V 4 on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K47 tok/s0.4 sPDF: MiniCPM V 4 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K71 tok/s0.4 sPDF: MiniCPM V 4 on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K60 tok/s0.4 sPDF: MiniCPM V 4 on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K81 tok/s0.2 sPDF: MiniCPM V 4 on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K103 tok/s0.2 sPDF: MiniCPM V 4 on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K112 tok/s0.2 sPDF: MiniCPM V 4 on RTX 4080 Super report
RTX 5070 Ti8-bit32K133 tok/s0.2 sPDF: MiniCPM V 4 on RTX 5070 Ti report
RTX 50808-bit32K141 tok/s0.1 sPDF: MiniCPM V 4 on RTX 5080 report
Radeon RX 7900 XTX8-bit32K114 tok/s0.4 sPDF: MiniCPM V 4 on Radeon RX 7900 XTX report
RTX 30908-bit32K138 tok/s0.2 sPDF: MiniCPM V 4 on RTX 3090 report
RTX 40908-bit32K147 tok/s0.1 sPDF: MiniCPM V 4 on RTX 4090 report
Radeon AI PRO R97008-bit32K81 tok/s0.2 sPDF: MiniCPM V 4 on Radeon AI PRO R9700 report
RTX 50908-bit32K236 tok/s0.1 sPDF: MiniCPM V 4 on RTX 5090 report
2× RTX 30908-bit32K138 tok/s0.1 sPDF: MiniCPM V 4 on 2× RTX 3090 report
2× RTX 40908-bit32K147 tok/s0.1 sPDF: MiniCPM V 4 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K188 tok/s0.1 sPDF: MiniCPM V 4 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K141 tok/s0.1 sPDF: MiniCPM V 4 on RTX 6000 Ada report
L40S8-bit32K129 tok/s0.1 sPDF: MiniCPM V 4 on L40S report
2× RTX 50908-bit32K236 tok/s0.1 sPDF: MiniCPM V 4 on 2× RTX 5090 report
4× RTX 30908-bit32K138 tok/s0.1 sPDF: MiniCPM V 4 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K236 tok/s0.1 sPDF: MiniCPM V 4 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit32K23 tok/s5.3 sPDF: MiniCPM V 4 on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K23 tok/s5.3 sPDF: MiniCPM V 4 on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K23 tok/s5.3 sPDF: MiniCPM V 4 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K48 tok/s2.6 sPDF: MiniCPM V 4 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K48 tok/s2.6 sPDF: MiniCPM V 4 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K48 tok/s2.6 sPDF: MiniCPM V 4 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K79 tok/s1.3 sPDF: MiniCPM V 4 on M4 Max 128 GB report
M5 Max 128 GB8-bit32K87 tok/s0.4 sPDF: MiniCPM V 4 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K97 tok/s0.7 sPDF: MiniCPM V 4 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K97 tok/s0.7 sPDF: MiniCPM V 4 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K97 tok/s0.7 sPDF: MiniCPM V 4 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K128 tok/s0.2 sPDF: MiniCPM V 4 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K35 tok/s0.8 sPDF: MiniCPM V 4 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K35 tok/s0.8 sPDF: MiniCPM V 4 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K49 tok/s0.3 sPDF: MiniCPM V 4 on RTX Spark report
DGX Spark8-bit32K44 tok/s0.3 sPDF: MiniCPM V 4 on DGX Spark report
Dual DGX Spark8-bit32K44 tok/s0.2 sPDF: MiniCPM V 4 on Dual DGX Spark report
DGX Station GB3008-bit32K559 tok/s0.1 sPDF: MiniCPM V 4 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K236 tok/s0.1 sPDF: MiniCPM V 4 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K236 tok/s0.1 sPDF: MiniCPM V 4 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K236 tok/s0.1 sPDF: MiniCPM V 4 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K260 tok/s0.2 sPDF: MiniCPM V 4 on A100 80 GB SXM report
H100 SXM8-bit32K368 tok/s0.1 sPDF: MiniCPM V 4 on H100 SXM report
4× H200 NVL8-bit32K458 tok/s0.1 sPDF: MiniCPM V 4 on 4× H200 NVL report
8× H200 NVL8-bit32K458 tok/s0.1 sPDF: MiniCPM V 4 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-gguf
vLLM
vllm serve openbmb/MiniCPM-V-4

Questions

How much VRAM does MiniCPM V 4 need?

MiniCPM V 4 needs 2.3 GB for its weights at 4-bit, 3.2 GB with an 8K context and 4.0 GB with 32K. At 8-bit, 4.9 GB.

What is the minimum VRAM for MiniCPM V 4?

About 2.8 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?

4.1B parameters: 7.6 GB of weights at 16-bit, 2.3 GB at 4-bit.

Can MiniCPM V 4 run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 147 tokens a second.

More models