Can I Run Qwen 2.5 VL on 128 GB system RAM?

Superseded model. Qwen 2.5 VL has been superseded by Qwen 3. This page is kept for reference; the newer family is a better starting point. View Qwen 3 →

作者： Jakub Rusinowski · 最后更新： 2025年1月25日

Technically yes, but not recommended

It loads, but it is not worth running — Qwen 2.5 VL 72B Instruct at Q8_0 fits in 128 GB system RAM's 102.4 GB, yet the memory bandwidth limits it to ~0.8 tok/s (estimated), well below usable interactive speed.

Confidence: high · Recommended quantization: Q8_0 · Estimated speed: ~0.8 tok/s

Buy This HardwareRyzen AI Max+ 395 Laptop (Strix Halo, up to 128GB) — 96 GB VRAM · 120 W board power Deploy in the Cloud NowRunPod

or compare on Vast.ai

As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.

128 GB system RAM — what it gives a model

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

Qwen 2.5 VL on 128 GB system RAM: memory by quantization

Quant	Memory needed	Fits 102.4 GB?	Max context	Est. speed	Download
F16	150.3 GB	✗ No	—	—	146.8 GB
Q8_0	81.5 GB	✓ Yes	64K	~0.8 tok/s	78 GB
Q6_K	63.7 GB	✓ Yes	64K	~1.1 tok/s	60.2 GB
Q5_K_M	55.5 GB	✓ Yes	64K	~1.3 tok/s	52 GB
Q4_K_M	47.8 GB	✓ Yes	64K	~1.5 tok/s	44.3 GB
Q3_K_M	34.8 GB	✓ Yes	64K	~2.1 tok/s	31.3 GB
Q2_K	27.6 GB	✓ Yes	64K	~2.6 tok/s	24.1 GB

Which Qwen 2.5 VL sizes fit

Variant	Needs at Q4_K_M	Fits?	Est. speed
Qwen 2.5 VL 72B Instruct	47.8 GB	✓ Fits	~1.5 tok/s
Qwen 2.5 VL 7B Instruct	6.3 GB	✓ Fits	~12.5 tok/s

What to watch out for

At ~0.8 tok/s this loads but is too slow for interactive use — expect roughly 75 seconds per 60 tokens.
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 computed from this model's published attention configuration.

FAQ