Configure a rig →

Qwen72B params128k context

Qwen 2.5 VL 72B locally

Qwen 2.5 VL 72B is an open-weight LLM from the Qwen family with 72B parameters. Main use: multimodal image workflows. Detected minimum hardware: Quadro RTX 8000 (48 GB).

Check on my GPU →Hugging Face ↗

Technical facts

Parameters72B

Max context128k

Q4_K_M45.3 GB

Q5_K_M55.3 GB

Q880.5 GB

FP16160.9 GB

FamilyQwen

Last sync2026-05-12

Available quantizations

GGUF weights

Q4_K_M

45.3GB

Acceptable. Good compromise when VRAM is limited.

Q5_K_M

55.3GB

Good quality. Sweet spot for size and precision.

Q8

80.5GB

Near-FP16 quality. Comfortable for production.

FP16

160.9GB

Reference precision. Maximum quality, doubled VRAM.

Compatible GPUs

12 single-GPU

GPUs that can run Qwen 2.5 VL 72B on a single card, ranked by VRAM headroom.

Quadro RTX 8000

48 GB · Quadro RTX

45.3 / 48 GBtight · Q4

48 GB · RTX A (Ampere)

45.3 / 48 GBtight · Q4

48 GB · RTX Ada

45.3 / 48 GBtight · Q4

48 GB · Ampere DC

45.3 / 48 GBtight · Q4

48 GB · Lovelace DC

45.3 / 48 GBtight · Q4

48 GB · Lovelace DC

45.3 / 48 GBtight · Q4

Radeon Pro W7900

48 GB · Radeon Pro W

45.3 / 48 GBtight · Q4

Mac Mini M4 Pro (48GB)

48 GB · Mac Mini

45.3 / 48 GBtight · Q4

MacBook Pro 14 M4 Pro (48GB)

48 GB · MacBook Pro 14

45.3 / 48 GBtight · Q4

MacBook Pro 14 M4 Max (48GB)

48 GB · MacBook Pro 14

45.3 / 48 GBtight · Q4

MacBook Pro 16 M3 Max (48GB)

48 GB · MacBook Pro 16

45.3 / 48 GBtight · Q4

MacBook Pro 16 M4 Pro (48GB)

48 GB · MacBook Pro 16

45.3 / 48 GBtight · Q4

Recommended multi-GPU rigs

2x / 4x consumer GPUs

For Qwen 2.5 VL 72B at higher quantization or with more context, a multi-GPU rig gives more headroom.

4× RTX 2060 12GB

48 GB · RTX 20

45.3 / 48 GBtight · Q4

48 GB · RTX 20

45.3 / 48 GBtight · Q4

4× RTX 3060 12GB

48 GB · RTX 30

45.3 / 48 GBtight · Q4

4× RTX 3080 12GB

48 GB · RTX 30

45.3 / 48 GBtight · Q4

4× RTX 3080 Ti

48 GB · RTX 30

45.3 / 48 GBtight · Q4

48 GB · RTX 30

45.3 / 48 GBtight · Q4

2× RTX 3090 Ti

48 GB · RTX 30

45.3 / 48 GBtight · Q4

48 GB · RTX 40

45.3 / 48 GBtight · Q4

Recommended rig

4× RTX 2060 12GB

Qwen 2.5 VL 72B with Ubuntu, vLLM, Open WebUI and the model already downloaded.

Similar models

VRAM estimates: parameters x bits/8 plus margin. Real performance varies by engine, context length and batch size.
sync: 2026-05-12