Skip to content

MiniCPM Llama3 V 2 5 VRAM requirements

How much VRAM MiniCPM Llama3 V 2 5 needs
4-bit weights: 4.8 GBWith 8K context: 6.6 GB8-bit, 8K context: 10.3 GB
Parameters
8.5B
Longest context
8K
Released
May 2024
Strongest at
Text and images

How much VRAM MiniCPM Llama3 V 2 5 needs

Memory MiniCPM Llama3 V 2 5 needs at each precision and context length
PrecisionWeightsWith 8K
16-bit15.9 GB17.7 GB
8-bit8.4 GB10.3 GB
6-bit6.5 GB8.3 GB
5-bit5.7 GB7.5 GB
4-bit4.8 GB6.6 GB
3-bit3.9 GB5.7 GB
2-bit3.0 GB4.8 GB
1.6-bit1.6 GB3.5 GB

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

GPUs and Macs that run MiniCPM Llama3 V 2 5

The best way MiniCPM Llama3 V 2 5 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40605-bit8K32 tok/s1.1 sPDF: MiniCPM Llama3 V 2 5 on RTX 4060 report
RTX 50605-bit8K52 tok/s0.9 sPDF: MiniCPM Llama3 V 2 5 on RTX 5060 report
RTX 30705-bit8K52 tok/s0.8 sPDF: MiniCPM Llama3 V 2 5 on RTX 3070 report
RTX 3080 10 GB6-bit8K75 tok/s0.6 sPDF: MiniCPM Llama3 V 2 5 on RTX 3080 10 GB report
RTX 3060 12 GB8-bit8K29 tok/s1.3 sPDF: MiniCPM Llama3 V 2 5 on RTX 3060 12 GB report
Arc B5808-bit8K25 tok/s0.9 sPDF: MiniCPM Llama3 V 2 5 on Arc B580 report
RTX 40708-bit8K40 tok/s0.6 sPDF: MiniCPM Llama3 V 2 5 on RTX 4070 report
RTX 4070 Super8-bit8K40 tok/s0.5 sPDF: MiniCPM Llama3 V 2 5 on RTX 4070 Super report
RTX 50708-bit8K52 tok/s0.5 sPDF: MiniCPM Llama3 V 2 5 on RTX 5070 report
RTX 4060 Ti 16 GB8-bit8K23 tok/s0.8 sPDF: MiniCPM Llama3 V 2 5 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit8K36 tok/s0.7 sPDF: MiniCPM Llama3 V 2 5 on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit8K31 tok/s0.8 sPDF: MiniCPM Llama3 V 2 5 on Arc A770 16 GB report
Radeon RX 9070 XT8-bit8K42 tok/s0.5 sPDF: MiniCPM Llama3 V 2 5 on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit8K52 tok/s0.4 sPDF: MiniCPM Llama3 V 2 5 on RTX 4070 Ti Super report
RTX 4080 Super8-bit8K57 tok/s0.3 sPDF: MiniCPM Llama3 V 2 5 on RTX 4080 Super report
RTX 5070 Ti8-bit8K69 tok/s0.4 sPDF: MiniCPM Llama3 V 2 5 on RTX 5070 Ti report
RTX 50808-bit8K73 tok/s0.3 sPDF: MiniCPM Llama3 V 2 5 on RTX 5080 report
Radeon RX 7900 XTX8-bit8K61 tok/s0.7 sPDF: MiniCPM Llama3 V 2 5 on Radeon RX 7900 XTX report
RTX 30908-bit8K72 tok/s0.5 sPDF: MiniCPM Llama3 V 2 5 on RTX 3090 report
RTX 40908-bit8K77 tok/s0.2 sPDF: MiniCPM Llama3 V 2 5 on RTX 4090 report
Radeon AI PRO R97008-bit8K42 tok/s0.5 sPDF: MiniCPM Llama3 V 2 5 on Radeon AI PRO R9700 report
RTX 50908-bit8K129 tok/s0.2 sPDF: MiniCPM Llama3 V 2 5 on RTX 5090 report
2× RTX 30908-bit8K72 tok/s0.3 sPDF: MiniCPM Llama3 V 2 5 on 2× RTX 3090 report
2× RTX 40908-bit8K77 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit8K100 tok/s0.2 sPDF: MiniCPM Llama3 V 2 5 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit8K73 tok/s0.2 sPDF: MiniCPM Llama3 V 2 5 on RTX 6000 Ada report
L40S8-bit8K67 tok/s0.3 sPDF: MiniCPM Llama3 V 2 5 on L40S report
2× RTX 50908-bit8K129 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on 2× RTX 5090 report
4× RTX 30908-bit8K72 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit8K129 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit8K11 tok/s11 sPDF: MiniCPM Llama3 V 2 5 on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit8K11 tok/s11 sPDF: MiniCPM Llama3 V 2 5 on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit8K11 tok/s11 sPDF: MiniCPM Llama3 V 2 5 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit8K25 tok/s5.3 sPDF: MiniCPM Llama3 V 2 5 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit8K25 tok/s5.3 sPDF: MiniCPM Llama3 V 2 5 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit8K25 tok/s5.3 sPDF: MiniCPM Llama3 V 2 5 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit8K43 tok/s2.7 sPDF: MiniCPM Llama3 V 2 5 on M4 Max 128 GB report
M5 Max 128 GB8-bit8K48 tok/s0.8 sPDF: MiniCPM Llama3 V 2 5 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit8K54 tok/s1.5 sPDF: MiniCPM Llama3 V 2 5 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit8K54 tok/s1.5 sPDF: MiniCPM Llama3 V 2 5 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit8K54 tok/s1.5 sPDF: MiniCPM Llama3 V 2 5 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit8K74 tok/s0.5 sPDF: MiniCPM Llama3 V 2 5 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit8K18 tok/s1.5 sPDF: MiniCPM Llama3 V 2 5 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit8K18 tok/s1.5 sPDF: MiniCPM Llama3 V 2 5 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit8K24 tok/s0.5 sPDF: MiniCPM Llama3 V 2 5 on RTX Spark report
DGX Spark8-bit8K22 tok/s0.5 sPDF: MiniCPM Llama3 V 2 5 on DGX Spark report
Dual DGX Spark8-bit8K22 tok/s0.3 sPDF: MiniCPM Llama3 V 2 5 on Dual DGX Spark report
DGX Station GB3008-bit8K374 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit8K129 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit8K129 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit8K129 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit8K144 tok/s0.3 sPDF: MiniCPM Llama3 V 2 5 on A100 80 GB SXM report
H100 SXM8-bit8K218 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on H100 SXM report
4× H200 NVL8-bit8K286 tok/s0.1 sPDF: MiniCPM Llama3 V 2 5 on 4× H200 NVL report
8× H200 NVL8-bit8K286 tok/s0.1 sPDF: MiniCPM Llama3 V 2 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-Llama3-V-2_5-gguf
vLLM
vllm serve openbmb/MiniCPM-Llama3-V-2_5

Questions

How much VRAM does MiniCPM Llama3 V 2 5 need?

MiniCPM Llama3 V 2 5 needs 4.8 GB for its weights at 4-bit, 6.6 GB with an 8K context and 6.6 GB with 8K. At 8-bit, 10.3 GB.

What is the minimum VRAM for MiniCPM Llama3 V 2 5?

About 5.7 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 Llama3 V 2 5?

8.5B parameters: 15.9 GB of weights at 16-bit, 4.8 GB at 4-bit.

Can MiniCPM Llama3 V 2 5 run on an RTX 4090?

Yes. At 8-bit it holds 8K of context and writes about 77 tokens a second.

More models