Models / MiniMax M3.1 Flash / VRAM
How much VRAM do you need for MiniMax M3.1 Flash?
Last updated:
Weight math from the model card, plus every measured RigBase result for MiniMax M3.1 Flash. Leave extra room for context (KV cache).
Weight size by quant
| Quant | Bits/weight | Weights only |
|---|---|---|
| Q8_0 | 8.50 | 454.8 GB |
| Q6_K | 6.56 | 351.1 GB |
| Q5_K_M | 5.70 | 305.1 GB |
| Q4_K_M | 4.89 | 261.9 GB |
| Q3_K_M | 4.00 | 213.8 GB |
| Q2_K | 3.16 | 169.0 GB |
Bits-per-weight for GGUF rows: llama.cpp quantize README. Parameter count: Hugging Face model card: ~428B total, 23B activated per token (quantization-awar. We advise staying at Q4 or above for chat quality — see Is Q4 good enough?.
Measured speeds for MiniMax M3.1 Flash
| Hardware | Quant | Engine | tok/s |
|---|---|---|---|
| 2x Nvidia GB10 (unified, DGX Spark) · 256 GB | EXL3 SAGE 2.15 bpw + DSpark | vLLM-EXL3 TP2, 2 nodes, 1 stream | 30.4 ⚡ |
| Apple M3 Ultra (60-core GPU, unified) · 256 GB | oQ3e-mtp (MTP off) | oMLX v0.7.2 | 25.3 |
| 4x Nvidia GB10 (unified, DGX Spark) · 512 GB | NVFP4 + DSpark | vLLM TP4, 4 nodes, 1 stream | 22 ⚡ |
| 2x Nvidia GB10 (unified, DGX Spark) · 256 GB | EXL3 SAGE 2.15 bpw + DSpark | vLLM-EXL3 TP2, 2 nodes, 1 stream | 19.8 ⚡ |
⚡ = speculative decoding (MTP, DFlash…). Full list: best home rig for MiniMax M3.1 Flash.
Related
All MiniMax M3.1 Flash builds VRAM + KV calculator By VRAM size GGUF vs AWQ vs FP8 vs NVFP4 How much system RAM?
FAQ
How much VRAM do you need for MiniMax M3.1 Flash?
Weights alone are about 261.9 GB at Q4_K_M (428B × 4.89 bits ÷ 8). Add memory for the KV cache (context) and the runtime. The smallest-memory measured result in RigBase is 256 GB (Apple M3 Ultra (60-core GPU, unified)) at 25.3 tok/s. Going below Q4 usually hurts quality for chat; see the Q4 guide.
What is the fastest measured speed for MiniMax M3.1 Flash?
Apple M3 Ultra (60-core GPU, unified): 25.3 tok/s plain decode (oQ3e-mtp (MTP off), oMLX v0.7.2).
Does system RAM replace VRAM?
Only for overflow. When the whole model fits in GPU or unified memory, system RAM barely matters for decode speed. See the system-RAM guide.
Referral link: if you sign up with Google and add $10, Runpod gives you and RigBase bonus credit (terms). It never affects our data.