Skip to content

MiniCPM5 1B VRAM requirements

How much VRAM MiniCPM5 1B needs
4-bit weights: 625 MBWith 8K context: 1.5 GBWith 32K context: 2.1 GB8-bit, 8K context: 2.0 GB
Parameters
1.1B
Longest context
128K
Released
May 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM MiniCPM5 1B needs

Memory MiniCPM5 1B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit2.0 GB2.9 GB3.5 GB5.8 GB
8-bit1.1 GB2.0 GB2.6 GB4.8 GB
6-bit845 MB1.8 GB2.3 GB4.6 GB
5-bit733 MB1.6 GB2.2 GB4.5 GB
4-bit625 MB1.5 GB2.1 GB4.4 GB
3-bit504 MB1.4 GB2.0 GB4.2 GB
2-bit386 MB1.3 GB1.9 GB4.1 GB
1.6-bit213 MB1.1 GB1.7 GB4.0 GB

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

GPUs and Macs that run MiniCPM5 1B

The best way MiniCPM5 1B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit128K178 tok/s0.1 sPDF: MiniCPM5 1B on RTX 4060 report
RTX 50608-bit128K270 tok/s0.1 sPDF: MiniCPM5 1B on RTX 5060 report
RTX 30708-bit128K270 tok/s0.1 sPDF: MiniCPM5 1B on RTX 3070 report
RTX 3080 10 GB8-bit128K404 tok/s0.1 sPDF: MiniCPM5 1B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit128K226 tok/s0.1 sPDF: MiniCPM5 1B on RTX 3060 12 GB report
Arc B5808-bit128K180 tok/s0.1 sPDF: MiniCPM5 1B on Arc B580 report
RTX 40708-bit128K297 tok/s0.1 sPDF: MiniCPM5 1B on RTX 4070 report
RTX 4070 Super8-bit128K297 tok/s0.1 sPDF: MiniCPM5 1B on RTX 4070 Super report
RTX 50708-bit128K369 tok/s0.1 sPDF: MiniCPM5 1B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit128K187 tok/s0.1 sPDF: MiniCPM5 1B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit128K270 tok/s0.1 sPDF: MiniCPM5 1B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit128K209 tok/s0.1 sPDF: MiniCPM5 1B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit128K275 tok/s0.1 sPDF: MiniCPM5 1B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit128K369 tok/s0.1 sPDF: MiniCPM5 1B on RTX 4070 Ti Super report
RTX 4080 Super8-bit128K394 tok/s0.1 sPDF: MiniCPM5 1B on RTX 4080 Super report
RTX 5070 Ti8-bit128K452 tok/s0.1 sPDF: MiniCPM5 1B on RTX 5070 Ti report
RTX 50808-bit128K474 tok/s0.1 sPDF: MiniCPM5 1B on RTX 5080 report
Radeon RX 7900 XTX8-bit128K354 tok/s0.1 sPDF: MiniCPM5 1B on Radeon RX 7900 XTX report
RTX 30908-bit128K466 tok/s0.1 sPDF: MiniCPM5 1B on RTX 3090 report
RTX 40908-bit128K489 tok/s0.1 sPDF: MiniCPM5 1B on RTX 4090 report
Radeon AI PRO R97008-bit128K275 tok/s0.1 sPDF: MiniCPM5 1B on Radeon AI PRO R9700 report
RTX 50908-bit128K682 tok/s0.1 sPDF: MiniCPM5 1B on RTX 5090 report
2× RTX 30908-bit128K466 tok/s0.1 sPDF: MiniCPM5 1B on 2× RTX 3090 report
2× RTX 40908-bit128K489 tok/s0.1 sPDF: MiniCPM5 1B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K583 tok/s0.1 sPDF: MiniCPM5 1B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K474 tok/s0.1 sPDF: MiniCPM5 1B on RTX 6000 Ada report
L40S8-bit128K441 tok/s0.1 sPDF: MiniCPM5 1B on L40S report
2× RTX 50908-bit128K682 tok/s0.1 sPDF: MiniCPM5 1B on 2× RTX 5090 report
4× RTX 30908-bit128K466 tok/s0.1 sPDF: MiniCPM5 1B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K682 tok/s0.1 sPDF: MiniCPM5 1B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit128K88 tok/s1.0 sPDF: MiniCPM5 1B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit128K88 tok/s1.0 sPDF: MiniCPM5 1B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit128K88 tok/s1.0 sPDF: MiniCPM5 1B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit128K164 tok/s0.5 sPDF: MiniCPM5 1B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K164 tok/s0.5 sPDF: MiniCPM5 1B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K164 tok/s0.5 sPDF: MiniCPM5 1B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K238 tok/s0.3 sPDF: MiniCPM5 1B on M4 Max 128 GB report
M5 Max 128 GB8-bit128K253 tok/s0.1 sPDF: MiniCPM5 1B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K272 tok/s0.1 sPDF: MiniCPM5 1B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K272 tok/s0.1 sPDF: MiniCPM5 1B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K272 tok/s0.1 sPDF: MiniCPM5 1B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K321 tok/s0.1 sPDF: MiniCPM5 1B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K137 tok/s0.1 sPDF: MiniCPM5 1B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K137 tok/s0.1 sPDF: MiniCPM5 1B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K193 tok/s0.1 sPDF: MiniCPM5 1B on RTX Spark report
DGX Spark8-bit128K178 tok/s0.1 sPDF: MiniCPM5 1B on DGX Spark report
Dual DGX Spark8-bit128K178 tok/s0.1 sPDF: MiniCPM5 1B on Dual DGX Spark report
DGX Station GB3008-bit128K1,101 tok/s0.1 sPDF: MiniCPM5 1B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K682 tok/s0.1 sPDF: MiniCPM5 1B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K682 tok/s0.1 sPDF: MiniCPM5 1B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K682 tok/s0.1 sPDF: MiniCPM5 1B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K727 tok/s0.1 sPDF: MiniCPM5 1B on A100 80 GB SXM report
H100 SXM8-bit128K894 tok/s0.1 sPDF: MiniCPM5 1B on H100 SXM report
4× H200 NVL8-bit128K1,002 tok/s0.1 sPDF: MiniCPM5 1B on 4× H200 NVL report
8× H200 NVL8-bit128K1,002 tok/s0.1 sPDF: MiniCPM5 1B 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 ewin-reg/MiniCPM5-1B-Agentic-Tooluse-v3-GGUF
vLLM
vllm serve openbmb/MiniCPM5-1B
MLX (Mac)
mlx_lm.chat --model mlx-community/MiniCPM5-1B-OptiQ-4bit

Questions

How much VRAM does MiniCPM5 1B need?

MiniCPM5 1B needs 625 MB for its weights at 4-bit, 1.5 GB with an 8K context and 2.1 GB with 32K. At 8-bit, 2.0 GB.

What is the minimum VRAM for MiniCPM5 1B?

About 1.4 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 MiniCPM5 1B?

1.1B parameters: 2.0 GB of weights at 16-bit, 625 MB at 4-bit.

Can MiniCPM5 1B run on an RTX 4090?

Yes. At 8-bit it holds 128K of context and writes about 489 tokens a second.

More models