MiniCPM MoE 8x2B — Hardware Requirements & GPU Compatibility
Chat
MiniCPM MoE 8x2B is a 2B-parameter open language model from openbmb. It supports a context window of up to 4,096 tokens. At BF16 it needs about 5.05 GB of VRAM — see which GPUs and Macs can run it below.
3.1K downloads
46 likesSpecifications
- Publisher
- openbmb
- Parameters
- 2B
- Architecture
- MiniCPMForCausalLM
- Context Length
- 4,096 tokens
- Vocabulary Size
- 122,753
- Release Date
- 2024-09-07
Get Started
HuggingFace
How Much VRAM Does MiniCPM MoE 8x2B Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| BF16 | 16.00 | 5.0 GB | 5.8 GB | 4.00 GB | Brain floating point 16 — preferred for training |
Which GPUs Can Run MiniCPM MoE 8x2B?
BF16 · 5.0 GBMiniCPM MoE 8x2B (BF16) requires 5.0 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 7+ GB is recommended. Using the full 4K context window can add up to 0.8 GB, bringing total usage to 5.8 GB. 35 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti.
Runs great
— Plenty of headroomWhich Devices Can Run MiniCPM MoE 8x2B?
BF16 · 5.0 GB33 devices with unified memory can run MiniCPM MoE 8x2B, including NVIDIA DGX H100, NVIDIA DGX A100 640GB.
Runs great
— Plenty of headroomRelated Models
Frequently Asked Questions
- How much VRAM does MiniCPM MoE 8x2B need?
MiniCPM MoE 8x2B requires 5.0 GB of VRAM at BF16. Full 4K context adds up to 0.8 GB (5.8 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 2B × 16 bits ÷ 8 = 4 GB
KV Cache + Overhead ≈ 1 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 1.8 GB (at full 4K context)
VRAM usage by quantization
BF165.0 GBBF16 + full context5.8 GB- Can I run MiniCPM MoE 8x2B on a Mac?
MiniCPM MoE 8x2B requires at least 5.0 GB at BF16, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
- Can I run MiniCPM MoE 8x2B locally?
Yes — MiniCPM MoE 8x2B can run locally on consumer hardware. At BF16 quantization it needs 5.0 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is MiniCPM MoE 8x2B?
At BF16, MiniCPM MoE 8x2B can reach ~577 tok/s on AMD Instinct MI300X. On NVIDIA GeForce RTX 4090: ~130 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: AMD Instinct MI300X → 5300 ÷ 5.0 × 0.55 = ~577 tok/s
Estimated speed at BF16 (5.0 GB)
~577 tok/s~130 tok/s~431 tok/s~357 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of MiniCPM MoE 8x2B?
At BF16, the download is about 4.00 GB.
- Which GPUs can run MiniCPM MoE 8x2B?
35 consumer GPUs can run MiniCPM MoE 8x2B at BF16 (5.0 GB). Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 35 GPUs have plenty of headroom for comfortable inference.
- Which devices can run MiniCPM MoE 8x2B?
33 devices with unified memory can run MiniCPM MoE 8x2B at BF16 (5.0 GB), including Mac Mini M4 (16 GB), Mac Mini M4 (32 GB), Mac Mini M4 Pro (24 GB), Mac Mini M4 Pro (48 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.