Can I Run MiniMax M2.5 on 128 GB system RAM?

作者： Jakub Rusinowski · 最后更新： 2026年2月15日

Yes, but it is tight

Yes, but it is tight — MiniMax M2.5 230B at Q3_K_M needs about 102.3 GB of the 102.4 GB usable on 128 GB system RAM, leaving only ~0.1 GB before the runtime starts swapping. Expect ~11 tok/s (estimated), with room for about 8,192 tokens of context.

Confidence: low · Recommended quantization: Q3_K_M · Estimated speed: ~11 tok/s

Get a personalized upgrade path →

购买此硬件 NVIDIA DGX Spark (128GB) — 128 GB VRAM · 150 W board power 立即云端部署 RunPod

或在 Vast.ai 比较

作为亚马逊联盟成员，我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金，您无需额外付费。

128 GB system RAM — what it gives a model

Usable memory for models	102.4 GB
Memory bandwidth	90 GB/s

MiniMax M2.5 on 128 GB system RAM: memory by quantization

Quant	Memory needed	Fits 102.4 GB?	Max context	Est. speed	Download
F16	464.1 GB	✗ No	—	—	459.8 GB
Q8_0	248.6 GB	✗ No	—	—	244.3 GB
Q6_K	192.8 GB	✗ No	—	—	188.5 GB
Q5_K_M	167.3 GB	✗ No	—	—	162.9 GB
Q4_K_M	143.1 GB	✗ No	—	—	138.8 GB
Q3_K_M	102.3 GB	✓ Yes	8K	~11 tok/s	98 GB
Q2_K	79.9 GB	✓ Yes	32K	~13.1 tok/s	75.6 GB

What to watch out for

Only ~0.1 GB of headroom at Q3_K_M: a longer context or a second application can push this into swapping.
Q3_K_M is the only quantization that fits, and it is a heavily degraded one — expect noticeably worse output than the same model at Q4_K_M. A smaller model at Q4 is usually the better trade.
1 larger variant of MiniMax M2.5 does not fit and would need CPU offload or different hardware.
This model does not publish its full attention configuration, so the KV-cache share of these figures is inferred from its parameter count rather than computed exactly.
These figures assume CPU-only inference. Any discrete GPU, even an 8 GB one, will be several times faster for models that fit in its VRAM.

Recommended setup

llama.cpp (CPU build) or Ollama — both run without a GPU

How these numbers are calculated

Memory need = quantized weights + KV cache at 8,192 tokens (f16) + 0.8 GB runtime overhead.
102.4 GB of the 128 GB is treated as usable for model weights (80% — the rest is the OS and running applications).
DDR5-5600 dual channel at 89.6 GB/s peak. CPU decode is assumed to sustain 35% of that peak, because CPU inference is not purely bandwidth-bound — it also spends real time in compute and thread synchronisation. This figure is an assumption, not a fitted constant: no CPU measurement is in the calibration set.
CPU-only inference: no GPU is assumed. A GPU of any size will beat these figures substantially.
Throughput is a memory-bandwidth roofline estimate, not a measurement. It is labelled "estimated" everywhere it appears.
KV cache is inferred from the parameter count because this model does not publish its full attention configuration.

FAQ