Qwen3.6 guide
Qwen3.6 27B is a 27B-class local model with an estimated 17GB Q4 runtime footprint in this dataset. It is a clean fit on 24GB and 32GB presets, but long multimodal context can still reduce headroom.
| Hardware | Clean capacity | Q4 need | Status | Calculator |
|---|---|---|---|---|
| 6 GB VRAM entry GPU | 4.5 GB clean VRAM | 17 GB | RAM offload | Open calculator |
| 8 GB VRAM mainstream GPU | 6.5 GB clean VRAM | 17 GB | RAM offload | Open calculator |
| 10 GB VRAM older high-end GPU | 8.5 GB clean VRAM | 17 GB | RAM offload | Open calculator |
| 12 GB VRAM local agent GPU | 10.5 GB clean VRAM | 17 GB | RAM offload | Open calculator |
| 16 GB VRAM creator GPU | 14.5 GB clean VRAM | 17 GB | RAM offload | Open calculator |
| 24 GB VRAM homelab workstation | 22.5 GB clean VRAM | 17 GB | Runs locally | Open calculator |
| 32 GB VRAM Blackwell workstation | 30.5 GB clean VRAM | 17 GB | Runs locally | Open calculator |
| 48 GB VRAM workstation | 46.5 GB clean VRAM | 17 GB | Runs locally | Open calculator |
| Apple Silicon 32 GB unified memory | 26 GB unified | 17 GB | Runs locally | Open calculator |
This is a practical planning estimate, not a benchmark. Real memory use changes with backend, context length, KV cache, quantization file, drivers, and offloading settings.