Can Qwen3.6 27B run on the RTX 5070?
Yes. At 8K context the best fit is Q2_K, using about 11.1 GB of the 11.4 GB usable on the RTX 5070. Estimated speed ceiling: ~65 tokens/sec.
Better quality with offload: 4-bit (Q4_K_M)
Runs with offload 4-bit (Q4_K_M) at 8K context: 39 of 64 layers on the RTX 5070 (about 10.7 GB) and the other 25 in system RAM (about 6.3 GB). Needs a PC with at least 32 GB of system RAM. Estimated speed: 3.3–6.3 tokens/sec (slow but usable), versus a bandwidth ceiling of ~40 tok/s if it fit fully in VRAM.
llama-server -hf unsloth/Qwen3.6-27B-GGUF:Q4_K_M -c 8192 -ngl 39
Estimate, not a benchmark. The command downloads the GGUF from Hugging Face. The speed range spans dual-channel DDR4-3200 to DDR5-6000 system RAM; your CPU, RAM speed and llama.cpp version move it. If llama.cpp runs out of GPU memory, lower -ngl by 2-3; if VRAM is left over, raise it for more speed. Recent llama.cpp builds can also pick the split automatically. How we estimate offload.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) | 10.8 GB Tight | 11.1 GB Tight | 12.6 GB No | 18.6 GB No | 65 |
| 3-bit (Q3_K_M) • | 13.8 GB No | 14.0 GB No | 15.5 GB No | 21.5 GB No | — |
| 4-bit (Q4_K_M) • | 16.8 GB No | 17.0 GB No | 18.5 GB No | 24.5 GB No | — |
| 5-bit (Q5_K_M) • | 19.3 GB No | 19.5 GB No | 21.0 GB No | 27.0 GB No | — |
| 6-bit (Q6_K) • | 22.1 GB No | 22.4 GB No | 23.9 GB No | 29.9 GB No | — |
| 8-bit (Q8_0) • | 27.8 GB No | 28.0 GB No | 29.5 GB No | 35.5 GB No | — |
| 16-bit (F16/BF16) • | 51.2 GB No | 51.5 GB No | 53.0 GB No | 59.0 GB No | — |
• real GGUF file size from unsloth/Qwen3.6-27B-GGUF. Speed is a bandwidth ceiling (672 GB/s), not a benchmark. Methodology.
What fits instead
- Cheapest listed GPU that runs Qwen3.6 27B: RX 7900 XT (20 GB, launch MSRP $899) at Q4_K_M, fully in VRAM at 8K context.
- Largest popular model that fits the RTX 5070: KORMo Web 11B (11B) at Q5_K_M, about 9.3 GB at 8K context.
"Runs" and "fits" mean the whole model in GPU memory at 8K context (4-bit or better for the model pick). Prices are US launch MSRPs, not today's street prices. "Popular" means the 60 most-downloaded models we track.
More
Qwen3.6 27B on every GPU and Mac · Every model that runs on the RTX 5070 · Other models that fit 12 GB VRAM · Try other settings in the calculator
Guides: Best GPU for local LLMs by budget · Can I run gpt-oss, Llama 70B or Qwen3? · Mac vs PC for local LLMs
Data: Hugging Face Hub (Qwen/Qwen3.6-27B), NVIDIA specs. Model data last checked 2026-11-17; refreshed hourly.