MiroThinker 1.7 Mini — Hardware Requirements & GPU Compatibility
ChatFunctions
258 downloads
57 likesSpecifications
- Publisher
- miromind-ai
- Parameters
- 30.5B
- Architecture
- Qwen3MoeForCausalLM
- Context Length
- 262,144 tokens
- Vocabulary Size
- 151,936
- Release Date
- 2026-03-12
- License
- Apache 2.0
Get Started
HuggingFace
How Much VRAM Does MiroThinker 1.7 Mini Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| BF16 | 16.00 | 61.5 GB | 74.3 GB | 61.06 GB | Brain floating point 16 — preferred for training |
Which GPUs Can Run MiroThinker 1.7 Mini?
BF16 · 61.5 GBMiroThinker 1.7 Mini (BF16) requires 61.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 80+ GB is recommended. Using the full 262K context window can add up to 12.8 GB, bringing total usage to 74.3 GB. No single GPU has enough memory — multi-GPU or cluster setups are needed.
Which Devices Can Run MiroThinker 1.7 Mini?
BF16 · 61.5 GB8 devices with unified memory can run MiroThinker 1.7 Mini, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Studio M4 Max (64 GB).
Runs great
— Plenty of headroom
Decent
— Enough memory, may be tightRelated Models
Frequently Asked Questions
- How much VRAM does MiroThinker 1.7 Mini need?
MiroThinker 1.7 Mini requires 61.5 GB of VRAM at BF16. Full 262K context adds up to 12.8 GB (74.3 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 30.5B × 16 bits ÷ 8 = 61.1 GB
KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 13.2 GB (at full 262K context)
VRAM usage by quantization
BF1661.5 GBBF16 + full context74.3 GB- Can NVIDIA GeForce RTX 5090 run MiroThinker 1.7 Mini?
No — MiroThinker 1.7 Mini requires at least 61.5 GB at BF16, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
- Can I run MiroThinker 1.7 Mini on a Mac?
MiroThinker 1.7 Mini requires at least 61.5 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 MiroThinker 1.7 Mini locally?
Yes — MiroThinker 1.7 Mini can run locally on consumer hardware. At BF16 quantization it needs 61.5 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is MiroThinker 1.7 Mini?
At BF16, MiroThinker 1.7 Mini can reach ~47 tok/s on AMD Instinct MI300X. 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 ÷ 61.5 × 0.55 = ~47 tok/s
Estimated speed at BF16 (61.5 GB)
AMD Instinct MI300X~47 tok/sNVIDIA H100 SXM~36 tok/sAMD Instinct MI250X~29 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of MiroThinker 1.7 Mini?
At BF16, the download is about 61.06 GB.
- Which GPUs can run MiroThinker 1.7 Mini?
No single consumer GPU has enough VRAM to run MiroThinker 1.7 Mini at BF16 (61.5 GB). Multi-GPU or professional hardware is required.
- Which devices can run MiroThinker 1.7 Mini?
8 devices with unified memory can run MiroThinker 1.7 Mini at BF16 (61.5 GB), including Mac Pro M2 Ultra (192 GB), Mac Studio M2 Ultra (192 GB), Mac Studio M4 Max (128 GB), Mac Studio M4 Max (64 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.